Short hairpin RNAs (shRNAs)
`induce sequence-specific silencing
`in mammalian cells
`
`Patrick J. Paddison,1 Amy A. Caudy,1 Emily Bernstein,2,3 Gregory J. Hannon,1,2,4
`and Douglas S. Conklin2
`1Watson School of Biological Sciences, 2Cold Spring Harbor Laboratory, Cold Spring Harbor, New York 11724, USA;
`3Graduate Program in Genetics, State University of New York at Stony Brook, Stony Brook, New York 11794, USA
`
`RNA interference (RNAi) was first recognized in Caenorhabditis elegans as a biological response to exogenous
`double-stranded RNA (dsRNA), which induces sequence-specific gene silencing. RNAi represents a conserved
`regulatory motif, which is present in a wide range of eukaryotic organisms. Recently, we and others have
`shown that endogenously encoded triggers of gene silencing act through elements of the RNAi machinery to
`regulate the expression of protein-coding genes. These small temporal RNAs (stRNAs) are transcribed as short
`hairpin precursors (∼70 nt), processed into active, 21-nt RNAs by Dicer, and recognize target mRNAs via
`base-pairing interactions. Here, we show that short hairpin RNAs (shRNAs) can be engineered to suppress the
`expression of desired genes in cultured Drosophila and mammalian cells. shRNAs can be synthesized
`exogenously or can be transcribed from RNA polymerase III promoters in vivo, thus permitting the
`construction of continuous cell lines or transgenic animals in which RNAi enforces stable and heritable gene
`silencing.
`[Key Words: RNAi; gene silencing; miRNA; shRNA; siRNA]
`
`Received January 31, 2002; revised version accepted March 8, 2002.
`
`An understanding of the biological role of any gene
`comes only after observing the phenotypic consequences
`of altering the function of that gene in a living cell or
`organism. In many cases, those organisms for which con-
`venient methodologies for genetic manipulation exist
`blaze the trail toward an understanding of similar genes
`in less tractable organisms, such as mammals. The ad-
`vent of RNA interference (RNAi) as an investigational
`tool has shown the potential to democratize at least one
`aspect of genetic manipulation, the creation of hypomor-
`phic alleles, in organisms ranging from unicellular para-
`sites (e.g., Shi et al. 2000) to mammals (Svoboda et al.
`2000; Wianny and Zernicka-Goetz 2000).
`Although Caenorhabditis elegans has, for some time,
`been well developed as a forward genetic system, the
`lack of methodologies for gene replacement by homolo-
`gous recombination presented a barrier to assessing rap-
`idly the consequences of loss of function in known
`genes. In an effort to overcome this limitation, Mello and
`Fire (Fire et al. 1998), building on earlier studies (Guo and
`Kemphues 1995), probed the utility of antisense RNA as
`
`4Corresponding author.
`E-MAIL hannon@cshl.org; FAX (516) 367-8874.
`Article and publication are at http://www.genesdev.org/cgi/doi/10.1101/
`gad.981002.
`
`a method for suppressing gene expression in worms.
`Through these efforts, they found that double-stranded
`RNA (dsRNA) was much more effective than antisense
`RNA as an inducer of gene silencing. Subsequent studies
`have shown that RNAi is a conserved biological response
`that is present in many, if not most, eukaryotic organ-
`isms (for review, see Bernstein et al. 2001b; Hammond et
`al. 2001b).
`As a result of biochemical and genetic approaches in
`several experimental systems, the mechanisms underly-
`ing RNAi have begun to unfold (for review, see Bernstein
`et al. 2001b; Hammond et al. 2001b). These suggest the
`existence of a conserved machinery for dsRNA-induced
`gene silencing, which proceeds via a two-step mecha-
`nism. In the first step, the dsRNA silencing trigger is
`recognized by an RNase III family nuclease called Dicer,
`which cleaves the dsRNA into ∼21–23-nt siRNAs (small
`interfering RNAs). These siRNAs are incorporated into a
`multicomponent nuclease complex, RISC, which identi-
`fies substrates through their homology to siRNAs and
`targets these cognate mRNAs for destruction.
`Although it was clear from the outset that RNAi
`would prove a powerful tool for manipulating gene ex-
`pression in invertebrates, there were several potential
`impediments to the use of this approach in mammalian
`cells. Most mammalian cells harbor a potent antiviral
`response that is triggered by the presence of dsRNA viral
`
`948
`
`GENES & DEVELOPMENT 16:948–958 © 2002 by Cold Spring Harbor Laboratory Press ISSN 0890-9369/02 $5.00; www.genesdev.org
`
`Benitec - Exhibit 1017 - page 1
`
`

`
`replication intermediates. A key component of this re-
`sponse is a dsRNA-activated protein kinase, PKR, which
`phosphorylates EIF-2␣, inducing, in turn, a generalized
`inhibition of translation (for review, see Williams 1997;
`Gil and Esteban 2000). In addition, dsRNA activates the
`2⬘5⬘ oligoadenylate polymerase/RNase L system and re-
`presses I␬B. The ultimate outcome of this set of re-
`sponses is cell death via apoptosis.
`Therefore, it came as a welcome surprise that dsRNA
`could induce sequence-specific silencing in mammalian
`embryos, which apparently lack generalized responses to
`dsRNA (Svoboda et al. 2000; Wianny and Zernicka-
`Goetz 2000).
`Indeed, microinjection of dsRNA into
`mouse zygotes could specifically silence both exogenous
`reporters and endogenous genes to create anticipated
`phenotypes. Subsequently, these observations were ex-
`tended to embryonic cell lines, such as embryonic stem
`cells and embryonal carcinoma cells, which do not show
`generic translational repression in response to dsRNA
`(Billy et al. 2001; Yang et al. 2001; Paddison et al. 2002).
`However, restriction of conventional RNAi to these few
`embryonic and cell culture systems would place a sig-
`nificant limitation on the utility of this approach in
`mammals.
`Tuschl and colleagues first showed that short RNA
`duplexes, designed to mimic the products of the Dicer
`enzyme, could trigger RNA interference in vitro in Dro-
`sophila embryo extracts (Tuschl et al. 1999; Elbashir et
`al. 2001b,c). This observation was extended to mamma-
`lian somatic cells by Tuschl and coworkers (Elbashir et
`al. 2001a) and by Fire and colleagues (Caplen et al. 2001),
`who showed that chemically synthesized siRNAs could
`induce gene silencing in a wide range of human and
`mouse cell lines. The use of synthetic siRNAs to tran-
`siently suppress the expression of target genes is quickly
`becoming a method of choice for probing gene function
`in mammalian cells.
`Dicer, the enzyme that normally produces siRNAs in
`vivo, has been linked to RNA interference both through
`biochemistry and through genetics (Bernstein et al.
`2001a; Grishok et al. 2001; Ketting et al. 2001; Knight
`and Bass 2001). Indeed, C. elegans animals that lack
`Dicer are RNAi-deficient, at least in some tissues. How-
`ever, these animals also have additional phenotypic ab-
`normalities. Specifically, they are sterile and show a
`number of developmental abnormalities that typify al-
`terations in developmental timing. Indeed, the pheno-
`types of the Dicer mutant animals were similar to those
`previously observed for animals carrying mutations in
`the let-7 gene (Reinhart et al. 2000).
`The let-7 gene encodes a small, highly conserved RNA
`species that regulates the expression of endogenous pro-
`tein-coding genes during worm development. The active
`RNA species is transcribed initially as an ∼70-nt precur-
`sor, which is posttranscriptionally processed into a ma-
`ture ∼21-nt form (Reinhart et al. 2000). Both in vitro and
`in vivo data from C. elegans (Grishok et al. 2001; Ketting
`et al. 2001; Knight and Bass 2001) and human cells
`(Hutvagner et al. 2001) have pointed to Dicer as the en-
`zyme responsible for let-7 maturation and for the matu-
`
`Stable silencing by RNAi
`
`ration of a similar small RNA, lin-4 (Grishok et al. 2001).
`Thus, at least some components of the RNAi machinery
`respond to endogenously encoded triggers to regulate the
`expression of target genes.
`Recent studies have placed let-7 and lin-4 as the found-
`ing members of a potentially very large group of small
`RNAs known generically as micro-RNAs (miRNAs).
`Nearly 100 potential miRNAs have now been identified
`in Drosophila, C. elegans, and mammals (Lagos-Quin-
`tana et al. 2001; Lau et al. 2001; Lee and Ambros 2001).
`Although the functions of these diverse RNAs remain
`mysterious, it seems likely that they, like let-7 and lin-4,
`are transcribed as hairpin RNA precursors, which are
`processed to their mature forms by Dicer (Lee and Am-
`bros 2001; E. Bernstein, unpubl.).
`Since the realization that small, endogenously en-
`coded hairpin RNAs could regulate gene expression via
`elements of the RNAi machinery, we have sought to
`exploit this biological mechanism for the regulation of
`desired target genes. Here we show that short hairpin
`RNAs (shRNAs) can induce sequence-specific gene si-
`lencing in mammalian cells. As is normally done with
`siRNAs, silencing can be provoked by transfecting exog-
`enously synthesized hairpins into cells. However, silenc-
`ing can also be triggered by endogenous expression of
`shRNAs. This observation opens the door to the produc-
`tion of continuous cells lines in which RNAi is used to
`stably suppress gene expression in mammalian cells.
`Furthermore, similar approaches should prove effica-
`cious in the creation of transgenic animals and poten-
`tially in therapeutic strategies in which long-term sup-
`pression of gene function is essential to produce a desired
`effect.
`
`Results
`
`Short hairpin RNAs trigger gene silencing
`in Drosophila cells
`
`Several groups (Grishok et al. 2001; Hutvagner et al.
`2001; Ketting et al. 2001; Knight and Bass 2001) have
`shown that endogenous triggers of gene silencing, spe-
`cifically small temporal RNAs (stRNAs) let-7 and lin-4,
`function at least in part through RNAi pathways. Spe-
`cifically, these small RNAs are encoded by hairpin pre-
`cursors that are processed by Dicer into mature, ∼21-nt
`forms. Moreover, genetic studies in C. elegans have
`shown a requirement for Argonaute-family proteins in
`stRNA function. Specifically, alg-1 and alg-2, members
`of the EIF2c subfamily, are implicated both in stRNA
`processing and in their downstream effector functions
`(Grishok et al. 2001). We have recently shown that a
`component of RISC, the effector nuclease of RNAi, is a
`member of the Argonaute family, prompting a model in
`which stRNAs may function through RISC-like com-
`plexes, which regulate mRNA translation rather than
`mRNA stability (Hammond et al. 2001a).
`We wished to test the possibility that we might retar-
`get these small, endogenously encoded hairpin RNAs to
`regulate genes of choice with the ultimate goal of sub-
`
`GENES & DEVELOPMENT
`
`949
`
`Benitec - Exhibit 1017 - page 2
`
`

`
`Paddison et al.
`
`verting this regulatory system for manipulating gene ex-
`pression stably in mammalian cell lines and in trans-
`genic animals. Whether triggered by long dsRNAs or by
`siRNAs, RNAi is generally more potent in the suppres-
`sion of gene expression in Drosophila S2 cells than in
`mammalian cells. We therefore chose this model sys-
`tem in which to test the efficacy of short hairpin RNAs
`(shRNAs) as inducers of gene silencing.
`Neither stRNAs nor the broader group of miRNAs
`that has recently been discovered form perfect hairpin
`structures. Indeed, each of these RNAs is predicted to
`contain several bulged nucleotides within their rather
`short (∼30-nt) stem structures. Because the position and
`character of these bulged nucleotides have been con-
`served throughout evolution and among at least a subset
`of miRNAs, we sought to design retargeted miRNA
`mimics to conserve these predicted structural features.
`Only the let-7 and lin-4 miRNAs have known mRNA
`targets (Wightman et al. 1993; Slack et al. 2000). In both
`cases, pairing to binding sites within the regulated tran-
`scripts is imperfect, and in the case of lin-4, the presence
`of a bulged nucleotide is critical to suppression (Ha et al.
`1996). We therefore also designed shRNAs that paired
`
`imperfectly with their target substrates. A subset of
`these shRNAs is depicted in Figure 1A.
`To permit rapid testing of large numbers of shRNA
`variants and quantitative comparison of the efficacy of
`suppression, we chose to use a dual-luciferase reporter
`system, as previously described for assays of RNAi in
`both Drosophila extracts (Tuschl et al. 1999) and mam-
`malian cells (Caplen et al. 2001; Elbashir et al. 2001a).
`Cotransfection of firefly and Renilla luciferase reporter
`plasmids with either long dsRNAs or with siRNAs ho-
`mologous to the firefly luciferase gene yielded an ∼95%
`suppression of firefly luciferase without effect on Renilla
`luciferase (Fig. 1B; data not shown). Firefly luciferase
`could also be specifically silenced by cotransfection with
`homologous shRNAs. Surprisingly, those shRNAs mod-
`eled most closely on the let-7 paradigm were the least
`effective inducers of silencing (data not shown). The in-
`clusion of bulged nucleotides within the shRNA stem
`caused only a modest reduction in potency; however, the
`presence of mismatches with respect to the target
`mRNA essentially abolished silencing potential. The
`most potent inhibitors were those composed of simple
`hairpin structures with complete homology to the sub-
`
`Figure 1. Short hairpins suppress gene
`expression in Drosophila S2 cells. (A) Se-
`quences and predicted secondary structure
`of representative chemically synthesized
`RNAs. Sequences correspond to positions
`112–134 (siRNA) and 463–491 (shRNAs) of
`Firefly luciferase carried on pGL3-Control.
`An siRNA targeted to position 463–485 of
`the luciferase sequence was virtually iden-
`tical to the 112–134 siRNA in suppressing
`(B) Exog-
`expression, but is not shown.
`enously supplied short hairpins suppress
`expression of the targeted Firefly lucifer-
`ase gene in vivo. Six-well plates of S2 cells
`were transfected with 250 ng/well of plas-
`mids that direct the expression of firefly
`and Renilla luciferase and 500 ng/well of
`the indicated RNA. Luciferase activities
`were assayed 48 h after transfection. Ra-
`tios of firefly to Renilla luciferase activity
`were normalized to a control transfected
`with an siRNA directed at the green fluo-
`rescent protein (GFP). The average of three
`independent experiments is shown; error
`bars indicate standard deviation. (C) Short
`hairpins are processed by the Drosophila
`Dicer enzyme. T7 transcribed hairpins
`shFfL22, shFfL29, and shFfS29 were incu-
`bated with (+) and without (−) 0–2-h Dro-
`sophila embryo extracts. Those incubated
`with extract produced ∼22-nt siRNAs,
`consistent with the ability of these hair-
`pins to induce RNA interference. A long
`dsRNA input (cyclin E 500-mer) was used
`as a control. Cleavage reactions were per-
`formed as described in Bernstein et al.
`(2001a).
`
`950
`
`GENES & DEVELOPMENT
`
`Benitec - Exhibit 1017 - page 3
`
`

`
`strate. Introduction of G-U basepairs either within the
`stem or within the substrate recognition sequence had
`little or no effect (Fig. 1A,B; data not shown). Similarly,
`varying either the loop size from ∼4 to 23 bases or the
`loop sequence (e.g., to mimic let-7) also proved neutral
`(data not shown).
`These results show that short hairpin RNAs can in-
`duce gene silencing in Drosophila S2 cells with potency
`similar to that of siRNAs (Fig. 1B). However, in our ini-
`tial observation of RNA interference in Drosophila S2
`cells, we noted a profound dependence of the efficiency
`of silencing on the length of the dsRNA trigger (Ham-
`mond et al. 2000). Indeed, dsRNAs of fewer than ∼200 nt
`triggered silencing very inefficiently. Silencing is initi-
`ated by an RNase III family nuclease, Dicer, that pro-
`cesses long dsRNAs into ∼22-nt siRNAs. In accord with
`their varying potency as initiators of silencing, long dsR-
`NAs are processed much more readily than short RNAs
`by the Dicer enzyme (Bernstein et al. 2001a). We there-
`fore tested whether shRNAs were substrates for the
`Dicer enzyme.
`We had noted previously that let-7 (Ketting et al. 2001)
`and other miRNAs (E. Bernstein, unpubl.) are processed
`by Dicer with an unexpectedly high efficiency as com-
`pared with short, nonhairpin dsRNAs. Similarly, Dicer
`efficiently processed shRNAs that targeted firefly lucif-
`erase, irrespective of whether they were designed to
`mimic a natural Dicer substrate (let-7) or whether they
`were simple hairpin structures (Fig. 1C). These data sug-
`gest that recombinant shRNAs can be processed by
`Dicer into siRNAs and are consistent with the idea that
`these short hairpins trigger gene silencing via an RNAi
`pathway.
`
`Short hairpin activated gene silencing
`in mammalian cells
`
`RNAi is developing into an increasingly powerful meth-
`odology for manipulating gene expression in diverse ex-
`perimental systems. However, mammalian cells contain
`several endogenous systems that were predicted to ham-
`per the application of RNAi. Chief among these is a
`dsRNA-activated protein kinase, PKR, which effects a
`general suppression of translation via phosphorylation of
`EIF-2␣ (Williams 1997; Gil and Esteban 2000). Activa-
`tion of these, and other dsRNA-responsive pathways,
`generally requires duplexes exceeding 30 bp in length,
`possibly to permit dimerization of the enzyme on its
`allosteric activator (e.g., Clarke and Mathews 1995).
`Small RNAs that mimic Dicer products, siRNAs, pre-
`sumably escape this limit and trigger specific silencing,
`in part because of their size. However, short duplex
`RNAs that lack signature features of siRNAs can effi-
`ciently induce silencing in Drosophila S2 cells but not in
`mammalian cells (A.A. Caudy, unpubl.). Endogenously
`encoded miRNAs may also escape PKR surveillance be-
`cause of their size but perhaps also because of the dis-
`continuity of their duplex structure. Given that shRNAs
`of <30 bp were effective inducers of RNAi in Drosophila
`
`Stable silencing by RNAi
`
`S2 cells, we tested whether these RNAs could also in-
`duce sequence-specific silencing in mammalian cells.
`Human embryonic kidney (HEK293T) cells were co-
`transfected with chemically synthesized shRNAs and
`with a mixture of firefly and Renilla luciferase reporter
`plasmids. As had been observed in S2 cells, shRNAs
`were effective inducers of gene silencing. Once again,
`hairpins designed to mimic let-7 were consistently less
`effective than were simple hairpin RNAs, and the intro-
`duction of mismatches between the antisense strand of
`the shRNA and the mRNA target abolished silencing
`(Fig. 2A; data not shown). Overall, shRNAs were some-
`what less potent silencing triggers than were siRNAs.
`Whereas siRNAs homologous to firefly luciferase rou-
`tinely yielded ∼90%–95% suppression of gene expres-
`sion, suppression levels achieved with shRNAs ranged
`from 80%–90% on average. As we also observe with siR-
`NAs, the most important determinant of the potency of
`the silencing trigger is its sequence. We find that roughly
`50% of both siRNAs and shRNAs are competent for sup-
`pressing gene expression. However, neither analysis of
`the predicted structures of the target mRNA nor analysis
`of alternative structures in siRNA duplexes or shRNA
`hairpins has proved of predictive value for choosing ef-
`fective inhibitors of gene expression.
`We have adopted as a standard, shRNA duplexes con-
`taining 29 bp. However, the size of the helix can be re-
`duced to ∼25 nt without significant loss of potency. Du-
`plexes as short as 22 bp can still provoke detectable si-
`lencing, but do so less efficiently than do longer
`duplexes. In no case do we observe a reduction in the
`internal control reporter (Renilla luciferase) that would
`be consistent with an induction of nonspecific dsRNA
`responses.
`The ability of shRNAs to induce gene silencing was
`not confined to 293T cells. Similar results were also ob-
`tained in a variety of other mammalian cell lines, includ-
`ing human cancer cells (HeLa), transformed monkey ep-
`ithelial cells (COS-1), murine fibroblasts (NIH 3T3), and
`diploid human fibroblasts (IMR90; Fig. 2; data not
`shown).
`
`Synthesis of effective inhibitors of gene expression
`using T7 RNA polymerase
`
`The use of siRNAs to provoke gene silencing is develop-
`ing into a standard methodology for investigating gene
`function in mammalian cells. To date, siRNAs have
`been produced exclusively by chemical synthesis (e.g.,
`Caplen et al. 2001; Elbashir et al. 2001a). However, the
`costs associated with this approach are significant, lim-
`iting its potential utility as a tool for investigating in
`parallel the functions of large numbers of genes. Short
`hairpin RNAs are presumably processed into active
`siRNAs in vivo by Dicer (see Fig. 1C). Thus, these may
`be more tolerant of terminal structures, both with re-
`spect to nucleotide overhangs and with respect to phos-
`phate termini. We therefore tested whether shRNAs
`could be prepared by in vitro transcription with T7 RNA
`polymerase.
`
`GENES & DEVELOPMENT
`
`951
`
`Benitec - Exhibit 1017 - page 4
`
`

`
`Paddison et al.
`
`Figure 2. Short hairpins function in mammalian cells. HEK 293T, HeLa, COS-1, and NIH 3T3 cells were transfected with plasmids
`and RNAs as in Figure 1 and subjected to dual luciferase assays 48 h posttransfection. The ratios of firefly to Renilla luciferase activity
`are normalized to a control transfected with an siRNA directed at the green fluorescent protein (GFP). The average of three indepen-
`dent experiments is shown; error bars indicate standard deviation.
`
`Transcription templates that were predicted to gener-
`ate siRNAs and shRNAs similar to those prepared by
`chemical RNA synthesis were prepared by DNA synthe-
`sis (Fig. 3A,C). These were tested for efficacy both in S2
`cells (data not shown) and in human 293 cells (Fig. 3B,D).
`Overall, the performance of the T7-synthesized hairpin
`or siRNAs closely matched the performance of either
`produced by chemical synthesis, both with respect to the
`magnitude of inhibition and with respect to the relative
`efficiency of differing sequences. Because T7 polymerase
`prefers to initiate at twin guanosine residues, however, it
`was critical to consider initiation context when design-
`ing in vitro transcribed siRNAs (Fig. 3B). In contrast,
`shRNAs, which are processed by Dicer (see Fig. 1C), tol-
`erate the addition of these bases at the 5⬘ end of the
`transcript.
`Studies in Drosophila embryo extracts indicate that
`siRNAs possess 5⬘ phosphorylated termini, consistent
`with their production by an RNase III family nuclease
`(Bernstein et al. 2001a; Elbashir et al. 2001b). In vitro,
`this terminus is critical to the induction of RNAi by
`synthetic RNA oligonucleotides (Elbashir et al. 2001c;
`Nykanen et al. 2001). Chemically synthesized siRNAs
`are nonphosphorylated, and enzymatic addition of a 5⬘
`phosphate group in vitro prior to transfection does not
`increase the potency of the silencing effect (A.A. Caudy,
`unpubl.). This suggests either that the requirement for
`phosphorylated termini is less stringent in mammalian
`
`952
`
`GENES & DEVELOPMENT
`
`cells or that a kinase efficiently phosphorylates siRNAs
`in vivo. RNAs synthesized with T7 RNA polymerase,
`however, possess 5⬘ triphosphate termini. We therefore
`explored the possibility of synthesizing siRNAs with T7
`polymerase followed by treatment in vitro with pyro-
`phosphatase to modify the termini to resemble those of
`siRNAs. Surprisingly, monophosphorylated siRNAs
`(data not shown) were as potent in inducing gene silenc-
`ing as transcription products bearing triphosphate ter-
`mini (Fig. 3B). This may suggest either that the require-
`ment for monophosphorylated termini is less stringent
`in mammalian cells or that siRNAs are modified in vivo
`to achieve an appropriate terminal structure.
`that both
`Considered together, our data suggest
`shRNAs and siRNA duplexes can be prepared by synthe-
`sis with T7 RNA polymerase in vitro. This significantly
`reduces the cost of RNAi in mammalian cells and paves
`the way for application of RNAi on a whole-genome
`scale.
`
`Transcription of shRNAs in vivo by RNA
`polymerase III
`
`Although siRNAs are an undeniably effective tool for
`probing gene function in mammalian cells, their sup-
`pressive effects are by definition of limited duration. De-
`livery of siRNAs can be accomplished by any of a num-
`
`Benitec - Exhibit 1017 - page 5
`
`

`
`Stable silencing by RNAi
`
`Figure 3.
`siRNAs and short hairpins transcribed in vitro suppress gene
`expression in mammalian cells. (A) Sequences and predicted secondary
`structure of representative in vitro transcribed siRNAs. Sequences corre-
`spond to positions 112–134 (siRNA) and 463–491 (shRNAs) of firefly lu-
`ciferase carried on pGL3-Control. (B) In vitro transcribed siRNAs suppress
`expression of the targeted firefly luciferase gene in vivo. HEK 293T cells
`were transfected with plasmids as in Figure 2. The presence of non-base-
`paired guanosine residues at the 5⬘ end of siRNAs significantly alters the
`predicted end structure and abolishes siRNA activity. (C) Sequences and
`predicted secondary structure of representative in vitro transcribed shR-
`NAs. Sequences correspond to positions 112–141 of firefly luciferase carried on pGL3-Control. (D) Short hairpins transcribed in vitro
`suppress expression of the targeted firefly luciferase gene in vivo. HEK 293T cells were transfected with plasmids as in Figure 2.
`
`ber of transient transfection methodologies, and both the
`timing of peak suppression and the recovery of protein
`levels as silencing decays can vary with both the cell
`type and the target gene (Y. Seger and E. Bernstein, un-
`publ.). Therefore, one limitation on siRNAs is the devel-
`
`opment of continuous cell lines in which the expression
`of a desired target is stably silenced.
`Hairpin RNAs, consisting of long duplex structures,
`have been proved as effective triggers of stable gene si-
`lencing in plants, in C. elegans, and in Drosophila (Ken-
`
`GENES & DEVELOPMENT
`
`953
`
`Benitec - Exhibit 1017 - page 6
`
`

`
`Paddison et al.
`
`nerdell and Carthew 2000; Smith et al. 2000; Tavernara-
`kis et al. 2000). We have recently shown stable suppres-
`sion of gene expression in cultured mammalian cells by
`continuous expression of a long hairpin RNA (Paddison
`et al. 2002). However, the scope of this approach was
`limited by the necessity of expressing such hairpins only
`in cells that lack a detectable PKR response. In principle,
`shRNAs could bypass such limitations and provide a
`tool for evoking stable suppression by RNA in mamma-
`lian somatic cells.
`To test this possibility, we initially cloned sequences
`encoding a firefly luciferase shRNA into a CMV-based
`expression plasmid. This was predicted to generate a
`capped, polyadenylated RNA polymerase II transcript in
`which the hairpin was extended on both the 5⬘ and 3⬘
`ends by vector sequences and poly(A). This construct
`was completely inert in silencing assays in 293T cells
`(data not shown).
`During our studies on chemically and T7-synthesized
`shRNAs, we noted that the presence of significant
`single-stranded extensions (either 5⬘ or 3⬘ of the duplex)
`reduced the efficacy of shRNAs (data not shown). We
`therefore explored the use of alternative promoter strat-
`egies in an effort to produce more defined hairpin RNAs.
`In particular, RNA polymerase III promoters have well-
`defined initiation and termination sites and naturally
`produce a variety of small, stable RNA species. Although
`many Pol
`III promoters contain essential elements
`within the transcribed region, limiting their utility for
`our purposes; class III promoters use exclusively non-
`transcribed promoter sequences. Of
`these,
`the U6
`snRNA promoter and the H1 RNA promoter have been
`well studied (Lobo et al. 1990; Hannon et al. 1991; Chong
`et al. 2001).
`By placing a convenient cloning site immediately be-
`hind the U6 snRNA promoter, we have constructed
`pShh-1, an expression vector in which short hairpins are
`harnessed for gene silencing. Into this vector either of
`two shRNA sequences derived from firefly luciferase
`were cloned from synthetic oligonucleotides. These
`were cotransfected with firefly and Renilla luciferase ex-
`pression plasmids into 293T cells. One of the two en-
`coded shRNAs provoked effective silencing of firefly lu-
`ciferase without altering the expression of the internal
`control (Fig. 4C). The second encoded shRNA also pro-
`duced detectable, albeit weak, repression. In both cases,
`silencing was dependent on insertion of the shRNA in
`the correct orientation with respect to the promoter (Fig.
`4C; data not shown). Although the shRNA itself is bilat-
`erally symmetric, insertion in the incorrect orientation
`would affect Pol III termination and is predicted to pro-
`duce a hairpin with both 5⬘ and 3⬘ single-stranded exten-
`sions. Similar results were also obtained in a number of
`other mammalian cell lines including HeLa, COS-1, NIH
`3T3, and IMR90 (Fig. 4; data not shown). pShh1-Ff1 was,
`however, incapable of effecting suppression of the lucif-
`erase reporter in Drosophila cells, in which the human
`U6 promoter is inactive (data not shown).
`As a definitive test of whether the plasmid-encoded
`shRNAs brought about gene silencing via the mamma-
`
`954
`
`GENES & DEVELOPMENT
`
`lian RNAi pathway, we assessed the dependence of sup-
`pression on an essential component of the RNAi path-
`way. We transfected pShh1-Ff1 along with an siRNA ho-
`to human Dicer. Figure 5 shows
`mologous
`that
`treatment of cells with Dicer siRNAs is able to com-
`pletely depress the silencing induced by pShh1-Ff1. Ad-
`dition of an unrelated siRNA had no effect on the mag-
`nitude of suppression by pShh1-Ff1 (data not shown). Im-
`portantly, Dicer siRNAs had no effect on siRNA-induced
`silencing of firefly luciferase (data not shown). These re-
`sults are consistent with shRNAs operating via an RNAi
`pathway similar to those provoked by stRNAs and long
`dsRNAs. Furthermore, it suggests that siRNA-mediated
`silencing is less sensitive to depletion of the Dicer en-
`zyme.
`The ultimate utility of encoded short hairpins will be
`in the creation of stable mutants that permit the study of
`the resulting phenotypes. We therefore tested whether
`we could create a cellular phenotype through stable sup-
`pression. Expression of activated alleles of the ras onco-
`gene in primary mouse embryo fibroblasts (MEFs) in-
`duces a stable growth arrest that resembles, as a terminal
`phenotype, replicative senescence (Serrano et al. 1997).
`Cells cease dividing and assume a typical large, flattened
`morphology. Senescence can be countered by mutations
`that inactivate the p53 tumor suppressor pathway (Ser-
`rano et al. 1997). As a test of the ability of vector-encoded
`shRNAs to stably suppress an endogenous cellular gene,
`we generated a hairpin that was targeted to the mouse
`p53 gene. As shown in Figure 6, MEFs transfected with
`pBabe-RasV12 fail to proliferate and show a senescent
`morphology when cotransfected with an empty control
`vector. As noted previously (Serrano et al. 1997), the ter-
`minally arrested state is achieved in 100% of drug-se-
`lected cells in culture by 8 d posttransfection. However,
`upon cotransfection of an activated ras expression con-
`struct with the pShh-p53, cells emerged from drug selec-
`tion that not only fail to adopt a senescent morphology
`but also maintain the ability to proliferate for a mini-
`mum of several weeks in culture (Fig. 6). These data
`strongly suggest that shRNA expression constructs can
`be used for the creation of continuous mammalian cell
`lines in which selected target genes are stably sup-
`pressed.
`
`Discussion
`
`The demonstration that short dsRNA duplexes can in-
`duce sequence-specific silencing in mammalian cells has
`begun to foment a revolution in the manner in which
`gene function is examined in cultured mammalian cells.
`These siRNAs (Elbashir et al. 2001a) mimic the products
`generated by Dicer (Bernstein et al. 2001a) in the initia-
`tion step of RNAi and presumably enter the silencing
`pathway without triggering nonspecific translational
`suppression via PKR. siRNAs can be used to examine the
`consequences of reducing the function of virtually any
`protein-coding gene and have proved effective in provok-
`ing relevant phenotypes in numerous somatic cell types
`from both humans and mice. However, a significant dis-
`
`Benitec - Exhibit 1017 - page 7
`
`

`
`Stable silencing by RNAi
`
`Figure 4. Transcription of functional shRNAs in vivo. (A) Schematic of the pShh1 vector. Sequences encoding shRNAs with between
`19 and 29 bases of homology to the targeted gene are synthesized as 60–75-bp double-stranded DNA oligonucleotides and ligated into
`an EcoRV site immediately downstream of the U6 promoter. (B) Sequence and predicted secondary structure of the Ff1 hairpin. (C) An
`shRNA expressed from the pShh1 vector suppresses luciferase expression in mammalian cells. HEK 293T, HeLa, COS-1, and NIH 3T3
`cells were transfected with reporter plasmids as in Figure 1, and pShh1 vector, firefly siRNA, or pShh1 firefly shRNA constructs as
`indicated. The ratios of firefly to Renilla luciferase activity were determined 48 h after tran