Gene Therapy (2006) 13, 503–508 & 2006 Nature Publishing Group All rights reserved 0969-7128/06 $30.00 www.nature.com/gt Induction of stable RNA interference in mammalian cells

Bryan R Cullen Center for Virology and Department of Molecular Genetics and Microbiology, Duke University Medical Center, Durham, NC, USA

Over the last years, RNA interference (RNAi) has become a and expression vectors that can facilitate the effective knock- widely used technique that permits the knock-down, and down of almost any vertebrate gene product in cultured cells hence functional analysis, of individual genes in vertebrate or in experimental animals. cells. However, the high failure rate of the RNA molecules Gene Therapy (2006) 13, 503–508. doi:10.1038/sj.gt.3302656; used in RNAi experiments continues to be a problem. In this published online 29 September 2005 paper, I describe a set of design criteria, experimental steps

Keywords: RNA interference; micro-RNAs; short hairpin RNAs; lentiviral vectors

Introduction which may be several thousand nucleotides in length, that is capped and polyadenylated16,17 (Figure 1). The RNA interference (RNAi) is an evolutionarily conserved mature miRNA forms part of one arm of an B80 nt mechanism that permits the selective post-transcrip- imperfect RNA stem–loop structure located within the tional downregulation of target genes in the cells of primary miRNA. This stem–loop structure is recognized metazoan eukaryotes, including humans.1 As such, by the nuclear microprocessor complex, consisting RNAi has enormous potential as not only an invaluable minimally of the RNase III enzyme Drosha and a tool in biological research and drug development but cofactor termed DGCR8, which cleaves the stem also as a possible approach to the in vivo inactivation B22 bp away from the terminal loop to release an of gene products linked to human disease progression B60 nt RNA hairpin intermediate called a pre-miRNA.18–21 and pathology. Pre-miRNAs are closely similar in structure to shRNAs, RNAi is mediated by B22 bp double-stranded RNA both of which are short RNA hairpins bearing an B2nt (dsRNA) molecules, bearing 2 nucleotide (nt) 30 over- 30 overhang that, in the case of the pre-miRNA, results hangs, termed short interfering RNA (siRNA) duplexes.1 from Drosha cleavage of the primary miRNA precursor These siRNA duplexes can be generated in situ by (Figure 1).18 Artificial shRNAs, in contrast, are normally cleavage of long dsRNAs, or artificial short hairpin RNAs directly transcribed by RNA polymerase III (polIII) and (shRNAs), by the cytoplasmic RNase III enzyme Dicer2–5 the 30 overhang results from polIII transcription termina- or can be directly introduced into cells by transfection.6 tion after the second T in a stretch of ‘T’ residues in the One strand of the siRNA duplex, termed the guide shRNA expression cassette (see below).2,3 strand, is then incorporated into the RNA-induced Both pre-miRNAs and shRNAs are then bound by a silencing complex (RISC),7–9 while the second, passenger complex consisting of the nuclear export factor Exportin strand, is released and degraded. Once incorporated into 5 (Exp5) and the GTP-bound form of its cofactor Ran.22,23 RISC, the siRNA guides this protein complex to mRNAs The Exp5/Ran–GTP complex, which recognizes small bearing highly complementary target sites, which can RNA stem–loops bearing short, B2nt 30 overhangs, then be cleaved by the action of the RISC-component transports pre-miRNAs and shRNAs to the cytoplasm Argonaut 2, leading to their degradation.8–11 where hydrolysis of the Ran-bound GTP moiety induces siRNAs appear functionally identical, at least in release of the RNA cargo. The pre-miRNAs and shRNAs mammalian cells, to endogenous B22 nt single-stranded now recruit the cytoplasmic RNase III enzyme Dicer, noncoding RNAs termed micro RNAs (miRNAs).12–14 which recognizes and binds short (B2 nt) 30 overhangs miRNAs are encoded within the host cell genome and present at the termini of dsRNA molecules.4,5 Dicer then over 200 are now known to exist in humans and mice.1,15 cleaves 21–22 nt away from the base of the RNA hairpin, Initially, miRNAs are transcribed by RNA polymerase II on both RNA strands, to remove the terminal loop and (polII) as part of a long primary miRNA transcript, generate the siRNA or miRNA duplex intermediate24 (Figure 1). Recent research has clarified the mechanism that Correspondence: Professor BR Cullen, Center for Virology and governs which strand of the duplex intermediate is then Department of Molecular Genetics and Microbiology, Box 3025, incorporated into RISC.25,26 Specifically, RISC incorpo- Duke University Medical Center, Durham, NC 27710, USA. 0 E-mail: [email protected] rates the RNA strand whose 5 -end is less tightly base- Received 1 June 2005; revised 15 August 2005; accepted 31 August paired. Thus, if one strand has stable G:C base pairs at its 2005; published online 29 September 2005 50-end, while the second has, for example, less stable A:U Expression of biologically active short hairpin RNAs BR Cullen 504

Figure 2 Incorporating pre-miRNA structural elements into shRNA designs facilitates their function. Shown is the predicted structure of the human pre-miR-1 precursor with the mature miRNA/siRNA shown in red and the passenger strand in blue. Dicer cleavage sites are indicated by arrows. The proposed shRNA precursor shown incorporates aspects of the pre-miR-1 structure as well as other advantageous design features, as discussed in detail in the text. Bases labeled as N are determined by the mRNA target site chosen, which is schematically represented at the bottom.

shRNAs should not contain extraneous inverted repeats that would interfere with the appropriate folding of the Figure 1 Short hairpin RNAs utilize the micro RNA (miRNA) shRNA hairpin. processing machinery. miRNAs are initially transcribed by polII as As shRNAs are effectively artificial orthologs of long pre-miRNAs and then processed by the nuclear RNase III enzyme Drosha to give the B60 nt pre-miRNA hairpin intermedi- endogenous pre-miRNAs (Figure 1), which are presum- ate. In contrast, shRNAs, which are functionally and structurally ably evolutionarily selected for efficiency and selectivity, equivalent to pre-miRNAs, are directly transcribed by polIII. The it is worthwhile to consider which attributes of subsequent nuclear export and cytoplasmic processing steps are pre-miRNAs might be usefully incorporated into indistinguishable for pre-miRNAs and shRNAs in human cells. See shRNAs.28,29 These considerations, together with exten- text for detailed discussion. sive analyses of artificial shRNA and siRNA function reported by several groups, suggest a number of design or G:U base pairs, or even a mismatch, then the latter criteria, as listed below. Throughout, I will use an shRNA design in which the 50 arm of the shRNA hairpin forms strand will be incorporated into RISC, while the former is 0 degraded. However, if the stability of base-pairing at the sense (passenger) strand, while the 3 arm forms the both ends of the duplex is similar, then selection of the antisense (guide) strand of the siRNA duplex intermediate. incorporated RNA strand becomes stochastic and fairly equivalent levels of each strand may enter RISC. For most, but not all, natural miRNAs only one strand of the shRNA transcription and processing duplex intermediate is generally detected in RISC. In the Transcription of shRNAs is normally performed using an case of artificial shRNAs, or siRNA duplexes, ensuring RNA polIII promoter such as the H1 or U6 promoter.2,3 that the correct (i.e. antisense) strand enters RISC is a Unlike many polIII-dependent promoters, H1 and U6 major design criterion, as described in more detail below. are compact promoters located 50 to the transcribed sequence. However, the U6 promoter strongly favors a shRNA design and target selection ‘G’ residue at the first position of the transcribed sequence, while the H1 promoter has a weak preference Optimization of the design of an shRNA to increase the for an ‘A’ residue. PolIII-mediated transcription termi- likelihood that it will effectively degrade its target nates after the second or, less commonly, third residue mRNA can be effectively subdivided into three steps. of a ‘TTTTT’ stretch, and the 30-terminal ‘UU’ of polIII- The first is ensuring efficient loading of RISC by the derived shRNAs is used to form the 30 B2 nt overhang antisense siRNA strand, the second is ensuring that the that is characteristic of pre-miRNAs (Figures 1 and 2). loaded siRNA has the potential to induce efficient target Both the preference for a specific residue at the +1 mRNA degradation, while the third is to select an mRNA position in the transcript, and the requirement for 2 ‘U’ target that is available for RISC binding. In addition, residues at the 30-end of artificial shRNA transcripts, it is important to avoid ‘off target’ effects of the shRNA influence both shRNA expression vector design and chosen by minimizing the potential of the shRNA mRNA target site selection (Figure 2). hairpin to activate host cell defense pathways, such As noted above, artificial shRNAs are transcribed in as the interferon response, that can be induced by the nucleus and processed in the cytoplasm by Dicer dsRNAs27 and by avoiding shRNAs that bear significant (Figure 1). Nuclear export of shRNAs and pre-miRNAs homology to other, irrelevant host mRNAs. Moreover, by Exp5 requires efficient recognition by the Exp5/

Gene Therapy Expression of biologically active short hairpin RNAs BR Cullen 505 Ran:GTP heterodimer. Fortunately, this heterodimer known about whether, or even if, siRNA primary requires only an RNA stem of X16 bp, a short 30 sequence, in and of itself, influences RISC function once overhang and a terminal loop of 46 nt.22,30 Similarly, loading has occurred, but evidence from computational efficient cleavage by Dicer appears to only require an analyses suggests an at least modest effect. Specifically, RNA stem–loop of sufficient size (i.e. a stem of analyses reported by Reynolds et al.32 suggest that an ‘A’ X19 bp)2,3 and a short 30 overhang.5 These characteristics residue at position 3 of the sense strand and a U at are therefore very easily incorporated into the shRNA position 10 of the sense strand enhance siRNA function design (Figure 2) and are a defining characteristic of significantly (Figure 2). In addition, avoiding a G at pre-miRNAs. position 13 of the sense strand appears to be helpful. Finally, there is evidence that runs of G’s and C’s Strand selection may be deleterious to function, while an excess of A’s As noted above, RISC selects the strand of the siRNA and U’s in, particularly, the 50 8 nt or so of the antisense duplex whose 50-end is less tightly base-paired.25,26 In the (guide) strand may facilitate siRNA function. These miRNA example shown in Figure 2, the 30 arm of the pre- latter design criteria may actually relate more to the next miR-1 precursor is incorporated into RISC as mature point. miR-1, while the 50 arm is excluded. Inspection shows that the 50-end of the excluded, 50 strand features two terminal G:C base-pairs and is tightly base-paired. In mRNA target site accessibility contrast, the 50-end of the included, 30 strand consists of a single, less stable A:U base-pair next to an A to G Unfortunately, even if your siRNA is optimally designed mismatch. Therefore, the artificial shRNA you design and efficiently incorporated into RISC, there is still no could contain two G:C base-pairs at the 50-end of the 50 guarantee that it will be effective. Current evidence strand and a weak base-pair (A:U or G:U) next to a suggests that this is due to the fact that RISC complexes mismatch at the 50-end of the 30, guide strand. This are not able to access mRNA target sites that are constrains the mRNA target site selected, as it will need occluded by protein binding or by even relatively weak to have G or C residues at the positions equivalent to the RNA secondary structure. Thus, Overhoff et al.33 have two 50 nucleotides of the passenger (sense) strand of presented compelling data demonstrating that local the shRNA (Figure 2). The mismatch at the 50-end of mRNA target secondary structure strongly inhibits the 30, guide (antisense) shRNA strand is introduced by siRNA efficiency, while Brown et al.34 used an elegant changing one position in the 50 passenger strand so it no approach to modulate artificially the folding of an longer matches the actual mRNA target sequence and is mRNA target site and thereby demonstrate that reduced therefore no longer complementary to the guide strand. mRNA folding leads to enhanced mRNA cleavage. Note that there are several ways to introduce a Perhaps the most compelling evidence showing that discrepancy in the base-pairing stability of the 50-ends mRNA secondary structure inhibits RNAi comes from of the siRNA duplex derived from the shRNA, and this Westerhout et al.,35 who selected a series of human is just one possible design approach. immunodeficiency virus type 1 (HIV-1) variants that had The approach suggested above for forcing incorpora- become resistant to an anti-viral siRNA upon serial tion of the antisense shRNA strand into RISC introduces passage in its presence. Remarkably, one HIV-1 variant a 1 nt bulge into the shRNA. Indeed, bulges are always proved to be highly resistant to a previously very present in natural pre-miRNAs and these mismatches effective siRNA, yet had no mutations within the chosen may fine-tune the cleavage sites used by Drosha and mRNA target site. The resistant HIV-1 variant did, Dicer and/or may preclude activation of dsRNA- however, contain a point mutation 7 nt 50 to the target responsive cellular signaling pathways, such as the site that, upon closer analysis, proved able to stabilize a interferon response, by cellular pre-miRNAs. Indeed, secondary structure that sequestered the viral mRNA recent preliminary data analyzing RNAi induced by target. Therefore, an otherwise fully functional siRNA siRNA duplexes that differ only in whether they form can be entirely ineffective when the mRNA target is a perfect duplex or instead contain a central, single- sequestered by another macromolecular interaction. nucleotide mismatch shows that, while both knock down While it remains quite difficult to predict whether a their specific mRNA target with equal efficiency, the given mRNA sequence is part of an RNA secondary siRNA duplexes bearing the mismatch cause far fewer structure, there has recently been significant progress in nonspecific effects (F Neipel, personal communication). developing computer algorithms that address this Therefore, although it has been previously reported that problem.33,36 For example, a potentially very useful efficient activation of the interferon response requires program for the identification of accessible target sites perfect dsRNAs of X30 bp in length,31 I strongly within mRNA molecules can be accessed at http:// recommend the inclusion of at least one additional bulge sfold.wadsworth.org.37 Beyond these more sophisticated within the shRNA design, as seen in native miRNAs, computational approaches, it is evident that G:C-rich instead of designing a perfect RNA helix. This is, of target sequences will be more likely to be sequestered course, again achieved by introducing a mutation into into RNA secondary structures than would A:U-rich the passenger (sense) strand of the shRNA (Figure 2), not sequences.32 While the ongoing difficulty in accurately into the guide strand. predicting whether a given mRNA target sequence is accessible means that even well-designed siRNAs fail at siRNA function a significant rate; our experience suggests that following While RISC loading of the shRNA antisense strand the shRNA design criteria listed above will give 480% should be efficient if the above rules are followed, how is knock-down of target mRNA-mediated protein expres- RISC cleavage efficiency ensured? Relatively little is sion by B1 out of three of the shRNAs tested.

Gene Therapy Expression of biologically active short hairpin RNAs BR Cullen 506 Testing an shRNA for effectiveness validation approach, it is therefore very simple to identify the shRNAs that are most effective at knocking Having selected an mRNA target sequence and designed down the targeted gene. an shRNA that is predicted to load RISC efficiently with an siRNA specific for that sequence, how do you Establishing stable RNAi in cultured cells determine whether you do get effective, that is, 480%, knock-down of the target mRNA and its encoded gene Having identified one or more highly effective shRNA product? Clearly, the first task is to construct an expression cassettes, the next question is how to produce expression vector, based on the U6 or H1 promoter, a stable knock-down of the target gene in the relevant which will transcribe the shRNA at high levels in cells. One possibility is to move the small H1-based transfected or transduced cells. A number of vectors shRNA expression cassette into a plasmid bearing a have been reported, but we favor the expression vector eukaryotic selectable marker and to then transfect and pSUPER, based on the H1 promoter, because of its subsequently select cells that express the marker. While convenient design.3 pSUPER contains convenient unique feasible, this approach is limited to cells that are fairly BglII and HindIII sites, the former located immediately 50 easy to stably transfect, that is, this approach would tend to the transcription start site, that allow synthetic DNA to work poorly in most primary cells. oligonucleotides encoding the B60 nt shRNA sequence, A better approach is to express the shRNA in the together with the 30-flanking ‘TTTTT’ transcription relevant cells using a retroviral40,41 or lentiviral vector.42–44 termination signal, to be readily inserted. Moreover, the These vectors can be readily generated at quite high resultant expression cassette is then flanked by several titers (X106/ml) by cotransfection into 293T cells unique restriction enzyme sites that allow it to be readily together with the relevant packaging constructs. More- moved into viral vectors (see below). over, they will readily infect most cells, including Once you have made a set of, say, five or more pSuper primary cells and (in the case of lentiviral vectors) vectors targeted to your gene of interest, you could simply nondividing cells, and will integrate into the target cell transfect each of these into an expressing cell line and use genome in a predictable manner. Finally, both retroviral Western analysis, or RT-PCR or Northern analysis, to see and, particularly, lentiviral vectors will generally give if gene knock-down occurs. This is, in my view, a poor you long-term, stable expression of the encoded strategy to validate the effectiveness of the encoded shRNA.42–44 siRNAs. One problem is that you may well be unsure as A range of retroviral or lentiviral vectors, bearing to the transfection efficiency of the cells you have markers that allow drug selection or selection by analyzed. If the target cells are primary cells, or fluorescence-activated cell sorting, have been described immortalized but differentiated cells such as B or T cells, and many of these appear to work well.40–44 One criterion the transfection efficiency you obtain may be fairly low. Of that we have noticed is important is that the vector note, a respectable 30% transfection efficiency could give should be self-inactivating (SIN), that is, the U3 region of younomorethana30%dropintargetmRNAorprotein the 30-long terminal repeat (LTR) in the vector plasmid expression levels even if the siRNA was 100% effective! should be largely deleted so that the 50-LTR promoter Secondly, the protein encoded by the targeted mRNA may generated during reverse transcription is inactive in the have a substantial half-life, so that only a modest drop in transduced cells. Moreover, the shRNA expression protein expression would be observed by 3 days post- cassette should be located 50 to any polII-dependent transfection even if the siRNA was 100% effective and promoter that is used to drive expression of the mRNA your transfection efficiency was also 100%! Conversely, encoding the selectable marker. The reason for this is that because RNAi in human cells is predominantly, or we have noted that the polIII-dependent H1 promoter is perhaps even entirely, a cytoplasmic process,38,39 quantita- subject to significant transcriptional interference from tion of the mRNA encoded by the targeted gene might polII-dependent promoters that are located 50. tend to underestimate the degree of knock-down obtained An effective lentiviral shRNA expression vector that with the shRNA tested, because the nuclear pool of the was developed in my laboratory,42 termed pNL-SIN-CMV, targeted mRNA would be largely unaffected even if the is shown in Figure 3 in a form expressing the blastocidin cytoplasmic mRNA pool was severely reduced. resistance marker (forms expressing green fluorescent To address these issues, we use a standard approach to protein are also available). This vector contains unique assess shRNA efficacy, which requires the construction of ClaIandXbaI restriction sites, present in the 30-LTR U3 an expression plasmid that encodes an epitope-tagged region, that allow the insertion of the entire shRNA form of the protein produced by the targeted gene. We expression cassette from pSuper derivatives. This not only then cotransfect 293T cells with this expression plasmid results in deletion of much of the lentiviral U3 region in and one of the pSuper derivatives, wait 2 days, lyse the transductants but also generates two copies of the H1- transfected cells and perform a Western analysis using an based shRNA expression cassette, one of which is located antiserum specific for the epitope tag. This approach 50 to the CMV promoter used to drive the selectable avoids the problems listed above because: (1) Cotrans- marker. This vector has been shown to give stable, readily fection is essentially 100% efficient, so that almost every detectable levels of siRNA expression in transduced cells cell that takes up the protein expression plasmid also and can be used to knock-down target genes for periods takes up the pSuper-based shRNA vector, even if only of at least several months.42 30% of the cells are actually transfected. (2) As there is no Although the focus of this review is on shRNA pre-existing pool of the epitope-tagged protein, protein expression in cultured cells, we note that others have half-life is not an issue. (3) As target protein expression is used lentiviral shRNA expression vectors to knock-down used as a read-out, the issue of a resistant, nuclear pool gene expression in vivo in mice.43,44 This knock-down of the targeted mRNA is not a concern. Using this was shown to be stable and ectopic expression of the

Gene Therapy Expression of biologically active short hairpin RNAs BR Cullen 507

Figure 3 Schematic of the NL-SIN-CMV lentiviral shRNA expression vector. This HIV-1-based lentiviral vector, shown here with the blastocidin resistance (BLR) marker, lacks the viral tat, rev, nef, env and vpu genes, but retains the viral gag, pol, vif and vpr genes as well as the Rev response element (RRE). In addition, almost all the 30-LTR U3 region has been deleted and replaced with unique ClaIandXbaI sites, which allow insertion of H1-promoter-based shRNA expression cassettes derived from pSuper. This expression cassette is duplicated during reverse transcription in susceptible infected cells. As shown previously,42 this vector is able to direct stable shRNA expression in transduced cells. encoded shRNAs did not have any nonspecific effects in response or target the same unknown but unrelevant these experimental animals. host mRNA transcript. To further support this hypo- thesis, you will need to show that expression of an irrelevant but functional shRNA (e.g. one targeted to green fluorescent protein, if this is not your selectable Confirming the specificity of RNAi marker) has no phenotype. These control experiments, phenotypes while not as definitive a demonstration of siRNA selectivity as the rescue experiment delineated above, Once you have transduced cells with a retroviral or are nevertheless generally viewed as an acceptable lentiviral expression vector encoding an shRNA specific alternate approach. Importantly, simply showing that for the gene of interest, you should see a substantial another, irrelevant or scrambled shRNA does not exert reduction in the level of expression of even stable the same phenotype as the experimental shRNA is not an proteins encoded by the targeted mRNA, if you wait a adequate control, as the irrelevant shRNA would not few days. If you do not see such a reduction, you should affect expression of a hypothetical mRNA that happened reconfirm the integrity of the lentiviral vector sequence to bear a sequence homologous to the one targeted by the and also confirm that you are indeed getting siRNA experimental shRNA. expression in the transduced cells, using Northern or primer extension analysis. We have not had a problem obtaining a good knock-down in transduced cells if good Conclusion expression of a previously validated siRNA is achieved. Assuming that you do get a good, stable knock-down Although RNAi is becoming an increasingly valuable of your target gene, and an interesting phenotype, how tool in both basic and applied biological research, and do you then show that this effect is specific, that is, not also has the potential to be used in vivo in the treatment due to the fortuitous knock-down of some other gene of disease, it remains difficult for researchers to design product whose mRNA bears a related sequence or due shRNAs or siRNA duplexes that consistently knock- to nonspecific activation of the host cell interferon down target gene expression to o20% of the level seen response? By far, the best way to validate the specificity in wild-type cells. In this review, I have attempted to of the observed phenotype is to reintroduce the targeted delineate a series of design criteria and technologies that gene in a form that is resistant to the siRNA used.45 That should allow readers to construct effective shRNA is, you can mutate the targeted region, if it is in the expression vectors specific for essentially any human mRNA open-reading frame, by taking advantage of the mRNA. Unfortunately, even armed with this informa- redundancy of the genetic code to introduce mutations tion, knock-down of a target gene by 480% will still only that do not alter the predicted protein product yet render be achieved with about one-third of the shRNAs tested. the encoded mRNA resistant to the shRNA used. While it would clearly be preferable to obtain efficient Alternately, if you have used a target sequence in the knock-down with every shRNA tested, a success rate of 30-untranslated region (30-UTR) of the mRNA of interest, one in three is acceptable in most experimental settings, you can simply delete the natural 30-UTR in the cDNA although it does greatly complicate efforts to use viral expression vector you design. Once you reintroduce the shRNA expression libraries as a way of performing RNAi-resistant form of the targeted gene, by transfection genetic screens. It remains to be seen if novel principles or transduction into the knock-down cells, the observed and insights will lead to future shRNA design criteria phenotype should return to the previous, wild-type state that give efficient target gene knock-down with every despite ongoing shRNA expression. shRNA analyzed. Regardless, RNAi in general, and If the above rescue approach is not feasible, then you shRNAs in particular, promise to lead to a revolution in could knock-down the gene using one or, preferably, two our understanding of the biological roles of vertebrate additional shRNAs that target entirely distinct sequences genes in health and disease. The reagents described in on the mRNA and show that you observe the same this review, including pSuper and the pNL-SIN-CMV phenotype. This control is based on the premise that all lentiviral shRNA expression vector, are available upon three shRNAs would not activate the same nonspecific request.

Gene Therapy Expression of biologically active short hairpin RNAs BR Cullen 508 Acknowledgements 22 Lund E, Guttinger S, Calado A, Dahlberg JE, Kutay U. Nuclear export of microRNA precursors. Science 2004; 303: 95–98. I thank Dr Alexandra Scha¨fer for assistance with the 23 Yi R, Qin Y, Macara IG, Cullen BR. Exportin-5 mediates the figures used in this manuscript. This work was supported nuclear export of pre-microRNAs and short hairpin RNAs. Genes by Grant GM071408 from the National Institutes of Health. Dev 2003; 17: 3011–3016. 24 Zeng Y, Yi R, Cullen BR. Recognition and cleavage of primary microRNA precursors by the nuclear processing enzyme References Drosha. EMBO J 2005; 24: 138–148. 25 Khvorova A, Reynolds A, Jayasena SD. Functional siRNAs and 1 Tomari Y, Zamore PD. Perspective: machines for RNAi. Genes miRNAs exhibit strand bias. Cell 2003; 115: 209–216. Dev 2005; 19: 517–529. 26 Schwarz DS, Hutvagner G, Du T, Xu Z, Aronin N, Zamore PD. 2 Paddison PJ, Caudy AA, Bernstein E, Hannon GJ, Conklin DS. Asymmetry in the assembly of the RNAi enzyme complex. Cell Short hairpin RNAs (shRNAs) induce sequence-specific silen- 2003; 115: 199–208. cing in mammalian cells. Genes Dev 2002; 16: 948–958. 27 Bridge AJ, Pebernard S, Ducraux A, Nicoulaz AL, Iggo R. 3 Brummelkamp TR, Bernards R, Agami R. A system for stable Induction of an interferon response by RNAi vectors in expression of short interfering RNAs in mammalian cells. Science mammalian cells. Nat Genet 2003; 34: 263–264. 2002; 296: 550–553. 28 Zeng Y, Wagner EJ, Cullen BR. Both natural and designed micro 4 Bernstein E, Caudy AA, Hammond SM, Hannon GJ. Role for a RNAs can inhibit the expression of cognate mRNAs when bidentate ribonuclease in the initiation step of RNA interference. expressed in human cells. Mol Cell 2002; 9: 1327–1333. Nature 2001; 409: 295–296. 29 Boden D, Pusch O, Silbermann R, Lee F, Tucker L, Ramratnam B. 5 Zhang H, Kolb FA, Brondani V, Billy E, Filipowicz W. Human Enhanced gene silencing of HIV-1 specific siRNA using micro- Dicer preferentially cleaves dsRNAs at their termini without a RNA designed hairpins. Nucleic Acid Res 2004; 32: 1154–1158. requirement for ATP. EMBO J 2002; 21: 5875–5885. 30 Zeng Y, Cullen BR. Structural requirements for pre-microRNA 6 Elbashir SM, Harborth J, Lendeckel W, Yalcin A, Weber K, Tuschl binding and nuclear export by Exportin 5. Nucleic Acid Res 2004; T. Duplexes of 21-nucleotide RNAs mediate RNA interference in 32: 4776–4785. cultured mammalian cells. Nature 2001; 411: 494–498. 31 Manche L, Green SR, Schmedt C, Matthews MB. Interactions 7 Hammond SM, Bernstein E, Beach D, Hannon GJ. An RNA- between double-stranded RNA regulators and the protein directed nuclease mediates post-transcriptional gene silencing in kinase. Mol Cell Biol 1992; 12: 5238–5248. Drosophila cells. Nature 2000; 404: 293–295. 32 Reynolds A, Leake D, Boese Q, Scaringe S, Marshall WS, 8 Schwarz DS, Hutva´gner G, Haley B, Zamore PD. Evidence that Khvorova A. Rational siRNA design for RNA interference. Nat siRNAs function as guides, not primers, in the Drosophila and Biotech 2004; 22: 326–330. human RNAi pathways. Mol Cell 2002; 10: 537–548. 33 Overhoff M, Alken M, Far RK, Lemaitre M, Lebleu B, Sczakiel G 9 Martinez J, Patkaniowska A, Urlaub H, Lu¨ hrmann R, Tuschl T. et al. Local RNA target structure influences siRNA efficacy: a Single-stranded antisense siRNAs guide target RNA cleavage in systematic global analysis. J Mol Biol 2005; 348: 871–881. RNAi. Cell 2002; 110: 563–574. 34 Brown KM, Chu C, Rana TM. Target accessibility dictates the 10 Liu J, Carmell MA, Rivas FV, Marsden CG, Thomson JM, Song JJ potency of human RISC. Nat Struct Mol Biol 2005; 12: 469–470. et al. Argonaute2 is the catalytic engine of mammalian RNAi. 35 Westerhout EM, Ooms M, Vink M, Das AT, Berkhout B. HIV-1 Science 2004; 305: 1437–1441. can escape from RNA interference by evolving an alternative 11 Meister G, Landthaler M, Patkaniowska A, Dorsett Y, Teng G, structure in its RNA genome. Nucleic Acid Res 2005; 33: 796–804. Tuschl T. Human Argonaute2 mediates RNA cleavage targeted 36 Ding Y, Lawrence CE. A statistical sampling algorithm for by miRNAs and siRNAs. Mol Cell 2004; 15: 185–197. RNA secondary structure prediction. Nucleic Acid Res 2003; 31: 12 Doench JG, Petersen CP, Sharp PA. siRNAs can function as 7280–7301. miRNAs. Genes Dev 2003; 17: 438–442. 37 Ding Y, Chan CY, Lawrence CE. Sfold web server for statistical 13 Zeng Y, Yi R, Cullen BR. MicroRNAs and small interfering RNAs folding and rational design of nucleic acids. Nucleic Acid Res can inhibit mRNA expression by similar mechanisms. Proc Natl 2004; 32: W135–W141. Acad Sci USA 2003; 100: 9779–9784. 38 Zeng Y, Cullen BR. RNA interference in human cells is restricted 14 Hutva´gner G, Zamore PD. A microRNA in a multiple-turnover to the cytoplasm. RNA 2002; 8: 855–860. RNAi enzyme complex. Science 2002; 297: 2056–2060. 39 Robb GB, Brown KM, Khurana J, Rana TM. Specific and potent 15 Lagos-Quintana M, Rauhut R, Lendeckel W, Tuschl T. Identifica- RNAi in the nucleus of human cells. Nat Struct Mol Biol 2005; 12: tion of novel genes coding for small expressed RNAs. Science 133–137. 2001; 294: 853–858. 40 Barton GM, Medzhitov R. Retroviral delivery of small interfer- 16 Cai X, Hagedorn CH, Cullen BR. Human microRNAs are ing RNA into primary cells. Proc Natl Acad Sci USA 2002; 99: processed from capped, polyadenylated transcripts that can 14943–14945. also function as mRNAs. RNA 2004; 10: 1957–1966. 41 Brummelkamp TR, Bernards R, Agami R. Stable suppression of 17 Lee Y, Kim M, Han J, Yeom KH, Lee S, Baek SH et al. MicroRNA tumorigenicity by virus-mediated RNA interference. Cancer Cell genes are transcribed by RNA polymerase II. EMBO J 2004; 23: 2002; 2: 243–247. 4051–4060. 42 Lee M-TM, Coburn GA, McClure MO, Cullen BR. Inhibition of 18LeeY,AhnC,HanJ,ChoiH,KimJ,YimJet al. The nuclear RNase human immunodeficiency virus type 1 replication in primary III Drosha initiates microRNA processing. Nature 2003; 425: 415–419. macrophages by using Tat- or CCR5-specific small interfering RNAs 19 Han J, Lee Y, Yeom KH, Kim YK, Jin H, Kim VN. The Drosha– expressed from a lentivirus vector. JVirol2003; 77: 11964–11972. DGCR8 complex in primary microRNA processing. Genes Dev 43 Rubinson DA, Dillon CP, Kwiatkowski AV, Sievers C, Yang L, 2004; 18: 3016–3027. Kopinja J et al. A lentivirus-based system to functionally silence 20 Gregory RI, Yan KP, Amuthan G, Chendrimada T, Doratotaj B, genes in primary mammalian cells, stem cells and transgenic Cooch N et al. The microprocessor complex mediates the genesis mice by RNA interference. Nat Genet 2003; 33: 401–406. of microRNAs. Nature 2004; 432: 235–240. 44TiscorniaG,SingerO,IkawaM,VermaIM.Ageneralmethodfor 21 Denli AM, Tops BB, Plasterk RH, Ketting RF, Hannon GJ. gene knockdown in mice by using lentiviral vectors expressing Processing of primary microRNAs by the microprocessor small interfereing RNA. Proc Natl Acad Sci USA 2003; 100: 1844–1848. complex. Nature 2004; 432: 231–235. 45 Whither RNAi? Nat Cell Biol 2003; 5: 489–490.

Gene Therapy