On the art of identifying effective and
`specific siRNAs
`
`Yi Pei & Thomas Tuschl
`
`Small interfering RNAs (siRNAs) have been widely exploited for sequence-specific gene
`knockdown, predominantly to investigate gene function in cultured vertebrate cells,
`and also hold promise as therapeutic agents. Because not all siRNAs that are cognate
`to a given target mRNA are equally effective, computational tools have been developed
`based on experimental data to increase the likelihood of selecting effective siRNAs.
`Furthermore, because target-complementary siRNAs can also target other mRNAs
`containing sequence segments that are partially complementary to the siRNA, most
`computational tools include ways to reduce potential off-target effects in the siRNA
`selection process. Though these methods facilitate selection of functional siRNAs, they
`do not yet alleviate the need for experimental validation. This perspective provides a
`practical guide based on current wisdom for selecting siRNAs.
`
`The evolutionarily conserved processes whereby small
`double-stranded (ds)RNAs of distinct size and structure
`sequence-specifically suppress the expression of their
`target genes are referred to as RNA silencing or RNA
`interference (RNAi)1. Among the repertoire of known
`small RNAs, siRNAs mediate gene-specific silencing pri-
`marily via recognizing and inducing degradation of the
`mRNAs of targeted genes. Consequently, siRNAs have
`become one of the most valuable reagents to function-
`ally annotate genomes and possess great potential as
`therapeutics2–4.
`Shortly after the discovery that siRNA duplexes can
`specifically silence mammalian genes, it was thought that
`almost any target-complementary siRNA effectively and
`specifically silences its cognate target gene5. In practice,
`however, different siRNAs often manifest a spectrum of
`potency, and only a fraction of them are highly effective6.
`Small positional shifts along the target mRNA were suffi-
`cient to alter siRNA function in an apparently unpredict-
`able manner6–8. Moreover, siRNAs may nonspecifically
`target unrelated genes with only partial sequence-com-
`plementarity (off-target effects)9–13. Hence, it is critical
`to identify effective and specific siRNA sequences to per-
`form reliable gene-knockdown experiments.
`
`Initially, empirical rules had been proposed for
`siRNA selection, some of which were based on the
`first identified functional siRNAs5. The evolving
`understanding of the RNAi mechanism, together with
`statistical analyses of libraries of siRNAs with experi-
`mentally determined efficiency, led to computer-
`based approaches that increased the likelihood of
`identifying effective and specific siRNAs6,14,15. These
`tools, however, are not perfect. (i) Not every selected
`siRNA meets the desired thresholds of potency and
`specificity, so that experimental proof of downregu-
`lation of targeted mRNA or protein remains impor-
`tant, not even considering the evaluation of potential
`off-target effects. (ii) A substantial fraction of active
`siRNAs may be dismissed because the weighing of fac-
`tors influencing activity is complex and partly unde-
`fined6,9,16. Not surprisingly, experimental approaches
`to generate and identify effective siRNAs have been
`developed to complement rule-driven siRNA selec-
`tion strategies16–18.
`There are many excellent recent reviews covering the
`mechanism of RNAi19–22. Elements of this mechanism
`that are important for the selection of siRNA are sum-
`marized in Box 1.
`
`PERSPECTIVE
`
`©2006 Nature Publishing Group http://www.nature.com/naturemethods
`
`Howard Hughes Medical Institute, Laboratory of RNA Molecular Biology, The Rockefeller University, 1230 York Avenue, Box 186, New York,
`New York 10021, USA. Correspondence should be addressed to T.T. (ttuschl@rockefeller.edu).
`PUBLISHED ONLINE 23 AUGUST 2006; DOI:10.1038/NMETH911
`
`670 | VOL.3 NO.9 | SEPTEMBER 2006 | NATURE METHODS
`
`Alnylam Exh. (cid:20)(cid:19)(cid:23)(cid:19)
`
`

`

`PERSPECTIVE
`
`BOX 1 siRNA-MEDIATED GENE SILENCING IN MAMMALIAN CELLS: ELEMENTS OF
`THE MECHANISM
`
`RNAi is a gene-regulatory mechanism triggered by dsRNAs.
`siRNAs, which consist of duplexes of 21–23 nt RNAs that are
`base-paired with 2-nt 3′ overhangs, mimic intermediates of
`the natural processing of longer double-stranded RNA triggers
`by RNase III. The major products of RNase III processing are
`microRNAs (miRNAs), which are endogenous ~ 22-nt RNAs that
`repress gene expression by targeting mRNAs for cleavage or
`translational repression22,72.
`An siRNA is generally designed to be fully complementary
`to its target mRNA and is commonly a product of chemical
`synthesis. Though the natural processing intermediates carry
`5′ phosphates, the 5′ phosphate is typically omitted in chemical
`synthesis as a cellular kinase rapidly phosphorylates the siRNA
`once it is delivered to the cells86. Individual siRNAs or random
`siRNA pools can also be generated by enzymatic methods from
`digestion of longer dsRNAs65,87,88. Alternatively, siRNAs can be
`generated by processing of ectopically expressed short hairpin
`RNAs89–91.
`Naturally occurring siRNAs cognate to cellular or viral mRNAs
`have not been experimentally detected in mammals. Instead, the
`mammalian RNAi machinery appears to have been adapted solely
`for miRNA-mediated regulation of mRNAs containing miRNA
`
`binding sites predominantly located in the 3′ UTR92. Although
`the majority of miRNAs affect the stability and translation of
`mRNAs that are only partially complementary72,92, some miRNAs
`use, like siRNAs, near-perfect complementarity to cleave their
`targets93. The latter requirement might explain the evolutionarily
`conserved catalytic aspect of targeted mRNA degradation in
`mammalian RNAi.
`RNase III and/or other components of the RNAi machinery
`specifically recognize an siRNA duplex and selectively incorporate
`one of the siRNA strands into different RISCs, including the
`catalytic endonuclease-containing complex, which is responsible
`for the strong siRNA gene-knockdown effect19,20. The strand
`antisense to the targeted mRNA is often referred to as the guide
`strand, and its base-paired sense strand is known as the passenger
`strand, which is destroyed upon incorporation of the guide strand
`into RISC47,94,95 (Fig. 1). The catalytic RISC recognizes mRNAs
`containing perfect or near-perfect complementary sequence to
`the guide siRNA and cleaves the mRNAs at a site precisely 10 nt
`upstream of the nucleotide opposite the 5′-most nucleotide of
`the guide strand (Fig. 1). The mRNA fragments are subsequently
`degraded by cellular nucleases, resulting in knockdown of the
`expression of the corresponding genes.
`
`Here we provide a practical guide and an overview of the theo-
`retical basis for identification and selection of effective and specific
`siRNAs.
`
`A GUIDE FOR siRNA SELECTION
`Target mRNA analysis
`The selection of siRNAs against a gene of interest starts with an
`annotated target mRNA sequence, including its 5′ and 3′ untrans-
`lated regions (UTRs), splice, polymorphic and allelic variants.
`Because the coding sequence is the most reliable mRNA sequence
`information available, it is commonly targeted. The UTRs are gen-
`erally less well characterized, but can also be targeted with similar
`gene-knockdown efficiency8,23,24. Though it has often been recom-
`mended to avoid targeting sequences that contain known binding
`sites for mRNA-binding proteins, such as the exon-exon junction
`complex, there is no detailed experimental study available to assess
`the importance of this guideline.
`For practical reasons, selection of siRNAs is often carried out with
`additional constraints, for example identifying siRNAs that target (i)
`orthologs in more than one species or (ii) all possible splice variants
`of a gene.
`
`Database search for published and validated siRNAs
`Several databases archive experimentally tested siRNA sequences
`from the literature25–27. Additionally, validated siRNAs can be
`acquired from commercial resources (for example, the Silencer vali-
`dated siRNAs from Ambion and HP validated siRNAs from Qiagen).
`Some vendors, such as Ambion, Qiagen and Dharmacon also pro-
`vide predesigned siRNAs or custom siRNA design service. Though
`prevalidated reagents provide an excellent starting point, the user
`
`still has to examine whether these siRNAs are potent and specific to
`meet the needs28.
`If there are no matches to the target gene of interest in any of these
`databases or in the literature, it is advisable to select 3–5 candidate
`siRNAs using available guidelines and tools, and subsequently to
`validate the reagents.
`
`Selected algorithms and siRNA sequence selection tools
`Several siRNA sequence selection algorithms have been developed in
`recent years that rely on intrinsic sequence and stability features of
`functional siRNAs6,14,15,23,29–35. A smaller number of algorithms
`consider the secondary structure and accessibility of the targeted
`mRNA36–38. The approaches underlying these algorithms range
`from empirical observations to sophisticated machine learning.
`After the siRNA sequence selection from the target mRNA sequence,
`each candidate siRNA is examined for similarity to all other mRNA
`transcripts that might unintentionally be targeted at a genome-wide
`level. Most of the siRNA selection algorithms have been combined
`with a variant of such programs, and the more user-friendly tools are
`listed in Table 1 (for a more complete list, see ref. 28). The selected
`siRNAs can be custom synthesized from four siRNA-licensed reagent
`suppliers: Ambion, Dharmacon, Qiagen and Sigma Proligo.
`
`Prevalidation of siRNAs
`Because the determination of the precise level of gene knock-
`down for each siRNA is a demanding process, and the assays need
`to be adapted for newly targeted genes, reporter-based assays
`have been developed to accelerate the identification of potent
`siRNAs among various synthesized siRNAs. In these systems,
`plasmids, which carry the target sequence fused to a reporter
`
`NATURE METHODS | VOL.3 NO.9 | SEPTEMBER 2006 | 671
`
`©2006 Nature Publishing Group http://www.nature.com/naturemethods
`
`

`

`PERSPECTIVE
`
`gene and a control gene for normalization, are cointroduced
`into cells together with the target-specific siRNAs14,39,40. The
`dual-luciferase–based siCHECK system from Promega is widely
`used and provides a ranking for siRNA activity within 24 h. The
`reporter-based activity generally correlates well with the efficacy
`of depleting the endogenous target (our unpublished observa-
`tions). The prevalidated siRNAs can then be used to validate the
`depletion of the endogenous target mRNA, which is discussed in
`detail in an accompanying review41.
`
`CONSIDERATIONS FOR SELECTING EFFECTIVE AND
`SPECIFIC siRNAs
`Sequence asymmetry of siRNA duplexes
`It has been demonstrated that structurally symmetric (duplexes
`with symmetric 2-nucleotide (nt) 3′ overhangs) but primary
`sequence–asymmetric (different nucleotides on each end) siRNAs,
`from which the target-mRNA complementary guide strand has
`greater propensity to be assembled into the RNA-induced silenc-
`ing complex (RISC) than the passenger strand, show improved effi-
`cacy and specificity42,43 (Fig. 1). The same finding emerges from
`sequence analysis of miRNA precursors and largely explains the
`asymmetric accumulation of the majority of miRNAs42. The asym-
`metry is determined by the different sequence composition, and the
`consequent differences in thermodynamic stability and molecular
`dynamic behavior of the two base-paired ends of an siRNA duplex:
`the strand with the less stable 5′ end, owing to either weaker base-
`pairing or introduction of mismatches, is favorably or exclusively
`loaded into RISC44. The asymmetry rule has been implemented in
`many siRNA design algorithms by computing either the A⋅U base
`pair content or local free energy at both ends of an siRNA, followed
`by selection of the duplexes with less stable, (A+U)-enriched 5′ end
`on the guide strand20.
`
`Because the majority of miRNAs start with a 5′ uridine, it is also
`conceivable that 5′ uridine–specific interaction contributes to more
`effective RISC assembly and function beyond the thermodynamic
`contributions discussed here. Furthermore, miRNA duplexes con-
`tain an average of six non–Watson-Crick base pairs distributed over
`the entire miRNA length, whose contribution to RISC assembly and
`asymmetry has not been evaluated.
`
`siRNA duplex stability
`Most analyzed functional siRNAs had a low-to-medium G+C
`content ranging between 30% and 52% (refs. 6,31). It has been
`argued that too low G+C content may destabilize siRNA duplexes
`and reduce the affinity for target mRNA binding, whereas too high
`G+C content may impede RISC loading and/or cleavage-product
`release. Additionally, surveys of functional siRNAs revealed that sta-
`ble duplexes devoid of internal repeats or palindromes, which may
`form intrastrand secondary structures, were better silencers6,31,45.
`An equally likely explanation is that the secondary structure of the
`target mRNA, which mirrors the predicted guide siRNA secondary
`structure, interferes with targeting.
`Although the overall duplex stability is important, the center of
`the duplex (positions 9–14 on the guide strand) appears to prefer-
`entially have low internal stability31,42,46. It has recently been noticed
`that miRNAs and siRNAs assemble into RISC by different mecha-
`nisms; siRNAs require cleavage of the passenger strand for effective
`RISC assembly, whereas a mismatched RNase III–processed miRNA
`duplex does not require passenger strand cleavage47. It is conceiv-
`able that the central-duplex instability may influence how effectively
`and to what ratios the RISC complexes with different core compo-
`nents are loaded. Alteration of the structure and stability of siRNA
`duplexes can also be controlled by incorporation of chemically
`modified nucleotide analogs. The effects of modifications, however,
`
`Table 1 | Representative siRNA sequence selection web tools
`Tools
`URLs
`siDESIGN
`http://www.dharmacon.com/
`
`RNAi Designer
`
`BIOPREDsi
`
`https://rnaidesigner.invitrogen.com/
`rnaiexpress/
`http://www.biopredsi.org
`
`Whitehead siRNA Selection server
`
`http://jura.wi.mit.edu/bioc/siRNA
`
`siDE
`
`siSearch
`
`http://side.bioinfo.ochoa.fib.es/
`
`http://sisearch.cgb.ki.se/
`
`Sirna
`
`http://sfold.wadsworth.org/sirna.pl
`
`siRNA design software
`
`http://www.cs.hku.hk/~sirna
`
`672 | VOL.3 NO.9 | SEPTEMBER 2006 | NATURE METHODS
`
`Comments
`Scores and ranks candidate siRNAs based on thermodynamic and
`sequence-related criteria. BLAST search is conducted by default.
`Ranks candidate siRNAs using a primitive scoring system.
`BLAST search is automatic and the results are shown.
`An artificial neural network–based tool, which was trained with
`~2,500 experimentally assessed siRNAs. Analysis of genome-wide
`specificity is included.
`Offers flexibility in defining siRNA sequence patterns and
`selection of filter functions. Different properties of selected
`siRNAs are calculated, including thermodynamic values,
`polymorphisms are identified and the results of configurable
`BLAST search and filtering are shown. The user can sort the
`output in various ways and balance decisions.
`Developed for high-throughput applications of siRNAs using
`several published algorithms for efficacy prediction and a
`nonredundant database for specificity analysis.
`The kernel algorithm focuses primarily on energy features of
`effective siRNAs. Alternative algorithms are also implemented
`and integrated in the tool. siSearch is expandable to include
`newly discovered rules.
`Sequence selection tool, which incorporates the target
`accessibility in the evaluation. No specificity analysis.
`Candidate siRNAs proposed by various previously developed
`sequence selection tools are classified based on target
`accessibility.
`
`References
`6,79
`
`Proprietary
`
`14
`
`30
`
`80
`
`31
`
`36
`
`38
`
`©2006 Nature Publishing Group http://www.nature.com/naturemethods
`
`

`

`PERSPECTIVE
`
`antisense strand. The A or U at position 10 is at the cleavage site and
`may promote catalytic RISC-mediated passenger strand and substrate
`cleavage. Other sequence determinants may be involved in steps along
`the RNAi pathway, such as RISC loading54.
`In addition to the positional nucleotide preference, certain motifs
`are commonly avoided in chemically synthesized siRNA duplexes
`that could affect the synthesis yield, purification or the annealing of
`siRNA strands. Extended runs of altering G⋅C pairs (more than 7)32
`or runs of more than three guanines are sometimes avoided.
`Moreover, in light of the reports that certain siRNAs can activate
`immune response in a cell- and sequence-dependent manner55–57,
`it is a prudent measure to filter out siRNA sequences containing
`putative immunostimulatory motifs in either strand to minimize
`toxicities and nonspecific silencing effects, especially when siRNAs
`are selected for in vivo and therapeutic use. Alternatively, immuno-
`stimulatory side effects can be masked using chemically modified
`nucleotides56,58,59. It is uncertain if all immunostimulatory RNA
`motifs have yet been defined.
`
`siRNA length and asymmetric 3′ overhang
`Conventionally designed siRNAs are 21-mers with symmetric 2-nt
`3′ overhangs, representing the predominant processing intermediate
`in the RNAi pathway60,61. It was noticed early that 20–25 nt siRNA
`duplexes carrying 2-nt 3′ overhangs could reach similar gene silenc-
`ing efficiency in mammalian cell culture experiments5,62, and that
`expressed or synthetic hairpin RNAs of up to 29 base pairs in length
`also triggered effective gene silencing63. These observations were fol-
`lowed up more recently, testing synthetic 21–29 nt RNA duplexes
`with blunt ends, symmetric or asymmetric 2-nt overhangs. These
`siRNA precursor molecules can silence target genes with similar
`efficiency to conventional siRNA duplexes64–68. The longer dsRNAs
`appear to have the advantage that they can be transfected at lower
`concentrations than conventional siRNAs without loss of gene
`silencing. But they also appear to be more likely to induce nonspe-
`cific responses (including interferon induction) or mediate other
`effects on cell viability67,68.
`Short RNA duplexes composed of single-strands of different
`length (19/21, 21/23 or 25/27 nt formats) have also been shown to
`silence mRNAs effectively66,69,70. The single 2-nt 3′ overhang present
`in those duplexes at the 3′ end of the guide strand and its presumed
`
`OH 3′
`
`Passenger strand
`
`5′ P
`3′ OH
`
`OH 3′
`Passenger strand (preferentially destroyed)
`P 5′
`Guide strand (preferentially enters RISC)
`RISC assembly
`
`Passenger strand destroyed
`
`3′ OH
`
`P 5′
`
`Guide strand
`
`RISC proteins
`Target recognition
`
`3′ OH
`
`Target mRNA
`Guide strand
`
`P 5′
`
`Figure 1 | A scheme for siRNA-mediated gene silencing. The primary
`sequence asymmetry of duplex determines which strand is preferentially
`assembled into RISC.
`
`are dependent on the position and the sequence context, and general
`rules are not yet available2,8,48.
`It is interesting to mention that imperfectly paired siRNA duplexes
`composed of target mRNA–complementary and partially palin-
`dromic or partially complementary single-stranded siRNAs have
`also been used successfully49. These siRNA duplexes are solely com-
`posed of two fully target-complementary guide strands that are
`sufficiently complementary to each other to form stable duplexes
`with characteristic 3′ overhanging ends. The silencing efficiencies
`of guide-only siRNA duplexes are comparable to prototypical fully
`paired passenger-guide duplex siRNAs, even though guide-only
`siRNA duplexes may contain a substantial number of non–Watson-
`Crick and G⋅U wobble base pairs.
`
`©2006 Nature Publishing Group http://www.nature.com/naturemethods
`
`Target accessibility
`It has been argued that local secondary structures (short stem-loops)
`in target mRNAs might restrict the accessibility of RISC, and attenu-
`ate or abolish siRNA efficacy37,50–53. A major obstacle in assessing
`target-accessibility is the lack of tools that reliably predict mRNA sec-
`ondary structure, setting aside the fact that mRNA is present inside
`cells as ribonucleoprotein complex of unknown composition. Several
`algorithms have been developed, and filtering of potentially inac-
`cessible target sequences has been shown to
`improve functional siRNA selection36–38,51.
`
`5′ P
`
`1
`
`2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21
`
`Sequence characteristics
`Several sequence analyses of siRNAs have
`independently identified single nucleo-
`tide positional preferences, which we will
`summarize using the guide strand as refer-
`ence6,14,23,29,32–35 (Fig. 2): (i) U or A at posi-
`tion 1; (ii) C or G (C is more common) at
`position 19; (iii) A+U richness between posi-
`tions 1 and 7; (iv) A or U (A is more com-
`mon) at position 10; (v) other motifs that
`were overrepresented in one analysis but not
`others, such as a U at position 17. The first
`three sequence features correlate with the
`rule of thermodynamic asymmetry, and the
`preferred nucleotides on indicated positions
`may contribute to the bias for selection of
`
`3′ OH
`
`3′ OH
`
`21 20 19
`
`18 17 16 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1
`
`P 5′
`
`Guide strand
`
`miRNA-like
`recognition
`
`Cleavage-based recognition
`
`21 20 19 18 17 16 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1
`
`Target mRNA
`
`P 5′
`
`Guide strand
`
`Figure 2 | siRNA and target mRNA structures. (a) Standard siRNA duplex. (b) Target mRNA specificity.
`The cleavage site is indicated by scissors in the target mRNA. Target recognition and off-target
`activity can occur in two modes, the catalytic siRNA-guided cleavage reaction requiring extensive
`complementarity in the region surrounding the cleavage site (blue) and the miRNA-like destabilization
`of mRNAs requiring pairing of the siRNA 5′ end (green).
`
`NATURE METHODS | VOL.3 NO.9 | SEPTEMBER 2006 | 673
`
`

`

`significant difference between two alleles, which may be as little
`as a single nucleotide change stemming from mutation or poly-
`morphism83. Placing this sequence discrepancy in the center of an
`siRNA, at or near the RISC cleavage site seems to be best for discrim-
`inating between alleles8,76–78,83. In some cases, introducing an addi-
`tional mismatch at other positions in the siRNA may improve the
`allele specificity, as long as the mismatch is tolerated for cleavage83.
`A limitation of this approach is that the choice of siRNA is restricted,
`and the siRNA may not be sufficiently effective. In this respect, it is
`interesting to note that the introduction of a G⋅U wobble mismatch
`in the 5′ terminal siRNA-mRNA interaction increased the potency
`of some siRNAs75. The efficacy of silencing may also be increased by
`destabilizing base-pairing at the 5′ end of the guide strand following
`the asymmetry rule78.
`Alternatively, both alleles can be nondiscriminately silenced by an
`effective siRNA distant from the polymorphic site, accompanied by
`ectopic expression of the desired sequence-modified allele refractory
`to the siRNA84. Vectors that simultaneously express transgene and
`short hairpin RNAs have been developed85.
`
`OUTLOOK
`In summary, guidelines are available that increase the likelihood of
`identifying effective and specific siRNAs at the expense of elimi-
`nating many potentially functional and specific siRNAs. These
`guidelines assist in reducing the numbers of siRNAs that need to
`be experimentally validated to identify potent and specific siRNAs
`for a given target gene. As reagent manufacturers have recognized
`the need for constant validation of siRNA knockdown experiments
`and developed promising lines of reagents, effective siRNAs can be
`identified at a rapid pace and will soon lead to the ultimate goal of
`production of validated genome-wide siRNA libraries needed for
`high-throughput or individual gene silencing experiments.
`
`ACKNOWLEDGMENTS
`We apologize to authors whose works are not cited owing to space limitations.
`We thank C. Echeverri at Cenix Bioscience for valuable discussion. We also thank
`M. Landthaler, P. Landgraf, J. Brennecke and C. Rogler for critical reading of the
`manuscript. Y.P. is supported by the Ruth L. Kirschstein Fellowship from the US
`National Institutes of Health–National Institute of General Medical Sciences.
`
`COMPETING INTERESTS STATEMENT
`The authors declare competing financial interests (see the Nature Methods
`website for details).
`
`Published online at http://www.nature.com/naturemethods/
`Reprints and permissions information is available online at http://npg.
`nature.com/reprintsandpermissions/
`
`1. Meister, G. & Tuschl, T. Mechanisms of gene silencing by double-stranded
`RNA. Nature 431, 343–349 (2004).
`2. Dorsett, Y. & Tuschl, T. siRNAs: applications in functional genomics and
`potential as therapeutics. Nat. Rev. Drug Discov. 3, 318–329 (2004).
`3. Echeverri, C.J. & Perrimon, N. High-throughput RNAi screening in cultured
`cells: a user’s guide. Nat. Rev. Genet. 7, 373–384 (2006).
`4. Dykxhoorn, D.M., Palliser, D. & Lieberman, J. The silent treatment: siRNAs as
`small molecule drugs. Gene Ther. 13, 541–552 (2006).
`5. Elbashir, S.M., Harborth, J., Weber, K. & Tuschl, T. Analysis of gene function
`in somatic mammalian cells using small interfering RNAs. Methods 26, 199–
`213 (2002).
`6. Reynolds, A. et al. Rational siRNA design for RNA interference. Nat.
`Biotechnol. 22, 326–330 (2004).
`7. Holen, T., Amarzguioui, M., Wiiger, M.T., Babaie, E. & Prydz, H. Positional
`effects of short interfering RNAs targeting the human coagulation trigger
`Tissue Factor. Nucleic Acids Res. 30, 1757–1766 (2002).
`8. Harborth, J. et al. Sequence, chemical, and structural variation of small
`interfering RNAs and short hairpin RNAs and the effect on mammalian gene
`
`PERSPECTIVE
`
`interaction with the RNAi machinery may contribute to asymmet-
`ric RISC assembly. However, the comparison with conventional
`siRNAs is again complicated by the differences in length of the guide
`strand and the differences in strength of 5′ terminal guide strand
`base-pairing when the paired region is longer than in conventional
`siRNA duplexes.
`
`Specificity
`Each strand of an siRNA duplex, once assembled into RISC, can
`guide recognition of fully and partially complementary target
`mRNAs, referred to as on- and off-targets, respectively. Though
`sequence asymmetry can be used to bias passenger strand exclu-
`sion, chemical methods of preventing passenger-strand use have also
`been introduced (for example, Dharmacon’s ON-TARGET siRNA).
`For the purpose of this discussion, we will distinguish off-targets
`into two classes (Fig. 2): (i) those that share contiguous and centrally
`located sequence complementarity over more than half of the siRNA
`sequence somewhere within the mRNA sequence71, and (ii) those
`that show solely 6 or 7 nucleotides of perfect match preferentially
`in the 3′ UTRs with positions 2–7 or 2–8 (seed region) of the guide
`siRNA9,11,12. The latter interaction is the major driving force behind
`endogenous miRNA–target mRNA recognition20,72. Although the
`off-targets of the latter class are predominant, their actual number
`identified in microarray analyses was significantly smaller than the
`number of computationally predicted targets with sequence com-
`plementary to the seed region of the guide strand, suggesting that
`additional specificity determinants remain to be identified9,12.
`Furthermore, structural and biochemical studies showed that
`guide-strand position 1 and the nucleotides at the 3′ overhang (posi-
`tions 20 and 21) have little, if any, contribution to the specificity of
`target recognition, and that mismatches near the 5′ and 3′ ends can
`be tolerated for RISC-guided cleavage if the remaining pairing to the
`target was unperturbed73,74.
`To enforce specificity, the current strategy is to select siRNAs in
`which the strand(s) entering RISC has some mismatches to all unde-
`sired target mRNAs, especially their 3′ UTRs. Typically at least three
`mismatches are recommended between positions 2 and 19 and the
`mismatches near the 5′ end and in the center of the examined strand
`should be assigned higher significance11,71,75,76. In addition to the
`position, the identity of the sequence mismatches also influence
`specificity to a certain extent75,77,78.
`Presently most tools use blastn or Smith-Waterman algo-
`rithm to remove potential off-targeting siRNAs during the siRNA
`sequence selection process79. In addition to the search method,
`the quality and completeness of the selected genome-wide mRNA
`sequence database is also of high importance79–81. The current
`tools, however, cannot eliminate all the potential off-targets, espe-
`cially those that contain the short sequence segments comple-
`mentary to the seed region of the guide strand, and likely discard
`many potentially functional siRNAs9. While improved algorithms
`are awaited, position-specific chemical modification of the
`seed-sequence of the guide siRNA can be used to reduce off-tar-
`get effects82. It is therefore important to experimentally control
`off-target effects or to dilute the off-target effects beyond the detection
`limit by codelivering several different target-specific siRNAs10,41.
`
`Allele-specific gene silencing
`To take advantage of the sequence specificity of RNAi, a prerequi-
`site to achieve allele-specific gene silencing is to identify the most
`
`674 | VOL.3 NO.9 | SEPTEMBER 2006 | NATURE METHODS
`
`©2006 Nature Publishing Group http://www.nature.com/naturemethods
`
`

`

`PERSPECTIVE
`
`40. Malik, I., Garrido, M., Bahr, M., Kugler, S. & Michel, U. Comparison of test
`systems for RNAinterference. Biochem. Biophys. Res. Commun. 341, 245–253
`(2006).
`41. Cullen, B.R. Enhancing and confirming the specificity of RNAi experiments.
`Nat. Methods 3, 677–681 (2006).
`42. Khvorova, A., Reynolds, A. & Jayasena, S.D. Functional siRNAs and miRNAs
`exhibit strand bias. Cell 115, 209–216 (2003).
`43. Schwarz, D.S. et al. Asymmetry in the assembly of the RNAi enzyme complex.
`Cell 115, 199–208 (2003).
`44. Hutvagner, G. Small RNA asymmetry in RNAi: function in RISC assembly and
`gene regulation. FEBS Lett. 579, 5850–5857 (2005).
`45. Patzel, V. et al. Design of siRNAs producing unstructured guide-RNAs results in
`improved RNA interference efficiency. Nat. Biotechnol. 23, 1440–1444 (2005).
`46. Taxman, D.J. et al. Criteria for effective design, construction, and gene
`knockdown by shRNA vectors. BMC Biotechnol. 6, 7 (2006).
`47. Matranga, C., Tomari, Y., Shin, C., Bartel, D.P. & Zamore, P.D. Passenger-
`strand cleavage facilitates assembly of siRNA into Ago2-containing RNAi
`enzyme complexes. Cell 123, 607–620 (2005).
`48. Manoharan, M. RNA interference and chemically modified small interfering
`RNAs. Curr. Opin. Chem. Biol. 8, 570–579 (2004).
`49. Hossbach, M., Gruber, J., Osborn, M., Weber, K. & Tuschl, T. Gene silencing
`with siRNA duplexes composed of target-mRNA-complementary and partially
`palindromic or partially complementary single-stranded siRNAs. RNA Biology
`3 (2006).
`50. Cullen, B.R. Induction of stable RNA interference in mammalian cells. Gene
`Ther. 13, 503–508 (2006).
`51. Heale, B.S., Soifer, H.S., Bowers, C. & Rossi, J.J. siRNA target site secondary
`structure predictions using local stable substructures. Nucleic Acids Res. 33,
`e30 (2005).
`52. Schubert, S., Grunweller, A., Erdmann, V.A. & Kurreck, J. Local RNA target
`structure influences siRNA efficacy: systematic analysis of intentionally
`designed binding regions. J. Mol. Biol. 348, 883–893 (2005).
`53. Overhoff, M. et al. Local RNA target structure influences siRNA efficacy: a
`systematic global analysis. J. Mol. Biol. 348, 871–881 (2005).
`54. Mittal, V. Improving the efficiency of RNA interference in mammals. Nat. Rev.
`Genet. 5, 355–365 (2004).
`55. Judge, A.D. et al. Sequence-dependent stimulation of the mammalian innate
`immune response by synthetic siRNA. Nat. Biotechnol. 23, 457–462 (2005).
`56. Hornung, V. et al. Sequence-specific potent induction of IFN-α by short
`interfering RNA in plasmacytoid dendritic cells through TLR7. Nat. Med. 11,
`263–270 (2005).
`57. Marques, J.T. & Williams, B.R. Activation of the mammalian immune system
`by siRNAs. Nat. Biotechnol. 23, 1399–1405 (2005).
`58. Judge, A.D., Bola, G., Lee, A.C. & MacLachlan, I. Design of noninflammatory
`synthetic siRNA mediating potent gene silencing in vivo. Mol. Ther. 13,
`494–505 (2006).
`59. Snøve, O. Jr. & Rossi, J.J. Expressing short hairpin RNAs in vivo. Nat. Methods
`3, 689–695 (2006).
`60. Elbashir, S.M. et al. Duplexes of 21-nucleotide RNAs mediate RNA
`interference in cultured mammalian cells. Nature 411, 494–498 (2001).
`61. Elbashir, S.M., Lendeckel, W. & Tuschl, T. RNA interference is mediated by 21-
`and 22-nucleotide RNAs. Genes Dev. 15, 188–200 (2001).
`62. Caplen, N.J.,