OLIGONUCLEOTIDES 18:305–320 (2008)
`© Mary Ann Liebert, Inc.
`DOI: 10.1089/oli.2008.0164
`
`REVIEW
`
`Chemical Modification of siRNAs for In Vivo Use
`
`Mark A. Behlke
`
`Well over a hundred reports have been published describing use of synthetic small-interfering RNAs (siRNAs) in
`animals. The majority of these reports employed unmodifi ed RNA duplexes. While unmodifi ed RNA is the natural
`effector molecule of RNA interference, certain problems arise with experimental or therapeutic use of RNA duplexes in
`vivo, some of which can be improved or solved through use of chemical modifi cations. Judicious use of chemical modi-
`fi cations can improve the nuclease stability of an RNA duplex, decrease the likelihood of triggering an innate immune
`response, lower the incidence of off-target effects (OTEs), and improve pharmacodynamics. This review will examine
`studies that document the utility of various chemical modifi cations for use in siRNAs, both in vitro and in vivo, with
`close attention given to reports demonstrating actual performance in animal model systems.
`
`Introduction
`
`The year 2008 marks the 30th anniversary of the fi rst
`
`experiments that employed synthetic oligonucleotides
`to specifi cally alter expression of targeted genes. In this
`early pioneering work, antisense DNA oligonucleotides
`were used to prevent translation of Rous Sarcoma virus
`RNA (Stephenson and Zamecnik, 1978; Zamecnik and
`Stephenson, 1978). Soon thereafter, antisense RNA was
`also shown to work as an experimental tool to inhibit gene
`expression (Melton, 1985; Graessmann et al., 1991). The effi -
`cacy of these early attempts to interfere with gene expression
`were limited by the short half-life of unmodifi ed DNA or
`RNA molecules, which are quickly degraded by the nucle-
`ases present in serum or living cells. Synthetic nucleic acids
`can be chemically modifi ed in ways that decrease their sus-
`ceptibility to nuclease degradation and also changes their
`properties in ways that can be useful for in vivo applications,
`such as reducing their ability to trigger an innate immune
`response and improving their general pharmacodynamic
`properties. Hundreds of different modifi cations have been
`studied for utility in antisense applications (Matteucci, 1997;
`Manoharan, 2002; Jason et al., 2004). A variety of chemically
`modifi ed antisense compounds have been successfully used
`for in vivo experimental biological studies and over a dozen
`candidate compounds are in various stages of clinical test-
`ing for use as therapeutic agents for a variety of disease indi-
`cations (Crooke, 2004). So far, only a single antisense agent
`has been approved for clinical use by the US Food and Drug
`Administration (FDA) [Vitravene, a 21-mer phosphorothioate
`
`(PS) DNA oligonucleotide which was approved in 1998 to
`treat CMV retinitis by direct intravitreal injection (Grillone
`and Lanz, 2001; Jabs and Griffi ths, 2002)].
`While “antisense” technology was the fi rst method that
`employed synthetic nucleic acids to alter gene expression,
`a number of other methodologies have since emerged that
`employ oligonucleotides as tools to regulate gene expres-
`sion levels that work through very different mechanisms
`of action. These methods include ribozymes, aptamers, and
`most recently, RNA interference (RNAi). RNAi is a natural
`gene regulatory network wherein short double-stranded
`RNAs (dsRNAs) interfere with the expression of com-
`plementary genes (Hannon and Rossi, 2004; Meister and
`Tuschl, 2004; Mello and Conte, 2004). It is an ancient pathway
`that is present in plants, mammals, and even some fungi. In
`primitive systems, RNAi can be effectively triggered by the
`presence of long dsRNA species. In higher organisms, long
`dsRNAs also trigger innate immune responses and their use
`can be toxic. Long dsRNAs are processed into shorter spe-
`cies which are the actual effector molecules that trigger an
`RNAi response. These shorter species, called “small-inter-
`fering RNAs” (siRNAs), are functional in mammalian cells
`and synthetic siRNAs can be safely used to experimentally
`activate RNAi (Elbashir et al., 2001).
`RNAi operates at multiple different sites within a cell and
`can suppress gene expression at the DNA level by inhibit-
`ing transcription (in the nucleus), by degrading messenger
`RNA (mRNA) (in the cytoplasm), or by directly suppressing
`translation (in the cytoplasm). The biochemistry of RNAi is
`
`Integrated DNA Technologies, Inc., Coralville, Iowa.
`
`305
`
`Alnylam Exh. 1039
`
`Downloaded by REPRINTS DESK INC from www.liebertpub.com at 03/29/18. For personal use only.
`
`

`

`306
`
`BEHLKE
`
`complex and the interaction of many different proteins is
`required to achieve gene suppression. Functioning in the
`cytoplasm, a heterodimer complex comprising TRBP and
`the endoribonuclease Dicer processes dsRNAs (23-base
`length or longer) into 21–22-mer siRNAs which have 2-base
`3′-overhangs and 5′-phosphates. Argonaute 2 (Ago2) joins
`Dicer/TRBP and together with the siRNA forms the sim-
`plest functional RISC (RNA-Induced Silencing Complex).
`In RISC, the siRNA is passed from Dicer/TRBP to Ago2
`and one strand (the “passenger” strand) of the siRNA is
`degraded (or unwound and discarded) and the remaining
`single-stranded RNA (ssRNA) (the “guide” strand) remains
`associated with Ago2 and directs the sequence specifi city of
`silencing by RISC. Ago2 contains an RNase H like domain
`which functions to cleave one strand of an RNA:RNA duplex
`and is the catalytic component of RISC that degrades a target
`RNA. Typically, perfect or near perfect base pairing between
`the siRNA guide strand and the target mRNA is required
`for Ago2 cleavage to occur.
`Of the four Ago proteins in mammalian cells, only Ago2
`has nuclease activity. The other three Ago proteins can nev-
`ertheless still be active components of RISC and it appears
`that some of these (at least Ago1) are involved in mediating
`the translational suppression arm of RNAi. Although any
`RNA loaded into RISC can theoretically trigger the trans-
`lational suppression pathway, endogenous microRNAs
`(miRNAs) are natural regulatory molecules that employ
`this mechanism to regulate gene expression. Endogenous
`miRNAs are encoded in genomic DNA and are expressed as
`long hairpin transcripts in the nucleus. These long hairpins
`are processed by the endoribonuclease Drosha in the nu-
`cleus and are exported to the cytoplasm where processing is
`completed by Dicer/TRBP prior to loading into RISC. Unlike
`siRNAs, imperfect base pairing between the target RNA
`and a miRNA triggers the translational repression activity
`of RISC. Small-interfering RNAs and miRNAs can function
`in either pathway so that imperfect base pairing between a
`siRNA and a target mRNA can result in unintended trans-
`lational suppression of that target and can cause undesired
`side effects (so called “Off Target Effects”, or OTEs) (Doench
`et al., 2003; Saxena et al., 2003; Yekta et al., 2004; Bagga et al.,
`2005). More details about the biochemistry of RNAi can be
`found in recent reviews (Hannon and Rossi, 2004; Meister
`and Tuschl, 2004; Mello and Conte, 2004; Filipowicz et al.,
`2005; Sontheimer, 2005; Grewal and Elgin, 2007; Rana, 2007;
`Filipowicz et al., 2008; Hutvagner and Simard, 2008). This
`review will primarily focus on the chemistry of the siRNA
`effector molecules that activate the degradative RNAi
`pathway.
`RNAi has rapidly progressed beyond use as an in vitro
`research tool and hundreds of reports already have been
`published describing use in animals. A variety of pharma-
`ceutical and biotechnology companies are actively work-
`ing toward development of RNAi-based therapeutics and
`human patients have already received siRNA drugs in clin-
`ical trials (Behlke, 2006; de Fougerolles et al., 2007; Kim and
`Rossi, 2007). While many of the early in vivo studies were
`done using unmodifi ed RNAs, much of the current work
`relating to RNAi therapeutics involves use of chemically
`modifi ed compounds.
`
`The Rationale for Chemical Modifi cation
`
`Most siRNAs used in research today are made by chem-
`ical synthesis using phosphoramidite building blocks as
`single-stranded oligonucleotides and are annealed into dou-
`ble-stranded form. This approach permits incorporation of a
`wide variety of natural and artifi cial modifi cations into the
`siRNA that can help solve some of the problems associated
`with administration of synthetic nucleic acids into cells or
`animals. Some problems that can be addressed using chem-
`ical modifi cation include:
`
`1. Susceptibility to nuclease degradation
`2. Activation of the innate immune system (OTE)
`3. Unwanted participation in miRNA pathways (OTE)
`4. Cell uptake and pharmacokinetics.
`
`The precise choice of chemical modifi cations to employ
`can vary with the design of the siRNA used, specifi c
`sequence, intended application, and method of delivery.
`Most siRNAs used today for in vivo research applications
`are synthetic 21-mer RNA duplexes that mimic the design
`of natural siRNAs. Similarly, 21-mer siRNAs are currently
`the lead compounds for a number of clinical and preclini-
`cal RNAi drug development programs. Alternative designs
`are also in use for both research and drug development,
`including blunt 19-mers (Czauderna et al., 2003; Allerson
`et al., 2005; Prakash et al., 2005; Kraynack and Baker, 2006),
`blunt 25-mers (Chen et al., 2005), blunt 27-mers (Kim et al.,
`2005), and asymmetric 25/27-mers or 27/29-mers (Rose et
`al., 2005; Dore-Savard et al., 2008; Nishina et al., 2008). Some
`of these compounds directly load into RISC whereas others
`are substrates for Dicer and are processed into shorter spe-
`cies before RISC loading. The precise design of the siRNA
`employed can infl uence the choice or pattern of chemical
`modifi cations suitable for use.
`Site selection is critical to the performance of siRNA com-
`pounds and a large amount of work from many different
`groups has led to the development of excellent site selection
`and design criteria as well as computer assisted algorithms
`that facilitate this process (Pei and Tuschl, 2006; Peek and
`Behlke, 2007). These design rules and algorithms were all
`developed using data from studies performed using unmod-
`ifi ed siRNAs. It is important to note that the use of chemical
`modifi cations can alter the potency of a siRNA and frequently
`a chemically modifi ed siRNA will show lower potency
`than the unmodifi ed RNA version of that same sequence,
`especially when extensively modifi ed. Specifi cally, certain
`modifi cation patterns have been demonstrated to impair
`the ability of an RNA sequence to trigger an RNAi effect
`whereas other patterns have less of an impact on potency.
`These effects vary with sequence, such that a given modifi -
`cation pattern can be effective at one site yet reduce potency
`of a siRNA at a second site. Thus empiric testing is usually
`necessary to ensure that a modifi ed siRNA is effective when
`a known potent unmodifi ed siRNA is converted to a mod-
`ifi ed form. Alternatively, initial site screening can be done
`using only modifi ed duplexes, bypassing use of unmodifi ed
`RNA entirely. Design rules and site selection criteria specifi c
`for modifi ed siRNAs have not been reported.
`
`Downloaded by REPRINTS DESK INC from www.liebertpub.com at 03/29/18. For personal use only.
`
`

`

`CHEMICAL MODIFICATION OF SIRNAS FOR IN VIVO USE
`
`307
`
`Susceptibility to Nuclease Degradation
`
`Single-stranded nucleic acids are rapidly degraded in
`serum or inside cells. Double-stranded nucleic acids, includ-
`ing siRNAs, are more stable than their single-stranded coun-
`terparts, but are still degraded and must be protected from
`nuclease attack if use includes exposure to serum. Protection
`can be provided externally through use of a suitable deliv-
`ery tool (such as complexation with a nanoparticle or encap-
`sulation within a liposome) or intrinsically through use of
`nuclease resistant chemical modifi cation of the nucleic acid
`itself. The primary nuclease activity in serum is a proces-
`sive 3′-exonuclease which may be ERI-1 (THEX1) (Eder et
`al., 1991; Kennedy et al., 2004), making 3′ end stabilization
`particularly important. The single-stranded 3′-overhangs
`present in traditional 21-mer siRNA designs are particularly
`susceptible to degradation. In serum, nuclease attack prob-
`ably initiates at the 3′-overhangs and proceeds in a 3′ → 5′
`direction (Zou et al., 2008). Placement of an inverted-dT base
`or other non-nucleotide groups at this position can protect
`against exonuclease attack. Interestingly, an RNase A like
`activity has also been implicated in the degradation of siR-
`NAs in serum (Haupenthal et al., 2007; Turner et al., 2007)
`and inhibiting RNase A activity can improve stability of siR-
`NAs (Haupenthal et al., 2006). RNase A cleaves at pyrimidine
`bases in ssRNA so it is not entirely clear how degradation of
`double-stranded siRNAs proceeds; it is possible that transient
`transitions to single-stranded form (“breathing”) in these
`short duplexes might provide a suitable substrate for attack
`by this class of nuclease. Longer RNAs, such as 27-mer Dicer-
`substrate duplexes, appear to have slightly greater inherent
`stability to serum nucleases (Kubo et al., 2007, 2008).
`When introducing chemical modifi cations into a siRNA,
`it is important to recall that the RNA must interact with a
`number of different cellular proteins, many or all of which
`may be sensitive to changes in RNA structure caused by the
`modifying group. The simplest approach to increase nuclease
`stability is to directly modify the internucleotide phosphate
`linkage. Replacement of a non-bridging oxygen with sul-
`fur (PS), boron (boranophosphate), nitrogen (phosphorami-
`date), or methyl (methylphosphonate) groups will provide
`nuclease resistance and have all been used to help stabilize
`single-stranded antisense oligonucleotides. Figure 1 shows
`various modifi cations that improve nuclease stability and
`can be employed in siRNAs. Phosphoramidate and meth-
`ylphosphonate derivatives were extensively explored for use
`in antisense applications and were found to signifi cantly alter
`interactions between the nucleic acid and cellular enzymes,
`such as RNase H. Their use has not been systematically stud-
`ied for use in RNAi. Boranophosphate modifi ed DNA or RNA
`is resistant to nuclease degradation and the boron modifi ca-
`tion appears to be compatible with siRNA function; however,
`boranophosphates are not easily made using chemical synthe-
`sis (Hall et al., 2004; Hall et al., 2006). The PS modifi cation is
`easily made and has been extensively used to improve nucle-
`ase stability of both antisense oligonucleotides and siRNAs.
`Phosphorothioate modifi ed nucleic acids are sulfated polyan-
`ions that are “sticky” and can nonspecifi cally bind to a variety
`of cellular proteins, potentially causing unwanted side effects
`(Krieg and Stein, 1995). Nevertheless, this modifi cation can
`
`be safely used to improve stability of a siRNA (Amarzguioui
`et al., 2003; Braasch et al., 2003; Chiu and Rana, 2003; Harborth
`et al., 2003; Li et al., 2005; Choung et al., 2006). Restricting place-
`ment of PS-modifi ed bonds to the ends of the oligonucleotides
`will provide resistance to exonucleases while minimizing the
`overall PS content of the oligo, thereby limiting unwanted side
`effects. Given the long history of use of PS-modifi ed antisense
`oligonucleotides, the potential toxicity of this modifi cation is
`well understood and PS-modifi ed compounds can be safely
`administered (Levin, 1999).
`Modifi cation of the 2′-position of the ribose can indi-
`rectly improve nuclease resistance of the internucleotide
`phosphate bond and at the same time can increase duplex
`stability (Tm) and may also provide protection from immune
`activation. 2′-O-methyl RNA (2′OMe) is a naturally occur-
`ring RNA variant found in mammalian ribosomal RNAs
`and transfer RNAs. It is nontoxic and can be placed within
`either the S or AS strands of a siRNA (Amarzguioui et al.,
`2003; Chiu and Rana, 2003; Czauderna et al., 2003; Harborth
`et al., 2003; Choung et al., 2006). Heavy modifi cation with
`2′OMe RNA can reduce potency or completely inactivate a
`siRNA; alternating 2′OMe with RNA bases generally retains
`siRNA function while conferring signifi cant nuclease stabi-
`lization (Czauderna et al., 2003; Choung et al., 2006; Santel et
`al., 2006a; Santel et al., 2006b). 2′OMe RNA can be combined
`with other 2′-modifi cations which are not naturally occur-
`ring bases with good results.
`The 2′-fl uoro (2′-F) modifi cation is compatible with
`siRNA function and also helps stabilize the duplex against
`nuclease degradation. Incorporation of 2′-F at pyrimidine
`positions maintains siRNA activity in vitro (Allerson et al.,
`2005; Prakash et al., 2005; Kraynack and Baker, 2006) and in
`vivo (Layzer et al., 2004). The 2′-F modifi cation is even tol-
`erated at the site of Ago2 cleavage (Muhonen et al., 2007).
`The combined use of 2′-F pyrimidines with 2′OMe purines
`can results in RNA duplexes with extreme stability in serum
`and improved in vivo performance (Morrissey et al., 2005b).
`The 2′-O-(2-methoxyethyl) RNA (MOE) modifi cation has
`been extensively used in antisense oligonucleotides and
`confers signifi cant nuclease stability to an oligonucleotide
`as well as increases Tm. 2′-MOE residues can be incorporated
`into siRNAs much like 2′OMe or 2′-F (Prakash et al., 2005),
`however this modifi cation is not generally available for use.
`The 2′-fl uoro-β-D-arabinonucleotide (FANA) modifi cation
`has also shown promise in antisense oligonucleotide appli-
`cations and can also be placed in siRNAs. Substitution of
`FANA for RNA in the entire S-strand confers signifi cant sta-
`bilization to nucleases while maintaining functional potency
`of the duplex; however, the AS-strand is less tolerant of the
`FANA modifi cation (Dowler et al., 2006).
`Locked nucleic acids (LNAs) contain a methylene bridge
`which connects the 2′-O with the 4′-C of the ribose. The
`methylene bridge “locks” the sugar in the 3′-endo confor-
`mation, providing both a signifi cant increase in Tm as well
`as nuclease resistance. Extensive modifi cation of a siRNA
`with LNA bases generally results in decreased activity (even
`more so than 2′OMe); however, siRNAs with limited incor-
`poration retain functionality and offer signifi cant nuclease
`stabilization (Braasch et al., 2003; Grunweller et al., 2003;
`Elmen et al., 2005; Mook et al., 2007).
`
`Downloaded by REPRINTS DESK INC from www.liebertpub.com at 03/29/18. For personal use only.
`
`

`

`308
`
`BEHLKE
`
`Base
`
`H
`
`H
`OH
`
`O
`
`O
`
`HO
`
`H
`
`O
`
`H
`
`O
`
`P
`
`O–
`
`RNA
`
`Base
`
`H
`
`F
`
`H
`
`O
`
`O
`
`H
`
`H
`
`O
`
`P
`
`S–
`
`Phosphorothioate
`
`O
`
`Boranophosphate
`
`O
`
`LNA
`
`Base
`
`H
`
`O
`
`H
`
`CH3
`
`2′-O-Methyl
`
`Base
`
`H
`
`H
`
`O
`
`P
`
`O–
`
`F
`
`H
`
`OH
`
`O
`
`O
`
`H
`
`O
`
`H
`
`O
`
`P
`
`O–
`
`FANA
`
`Base
`
`2′-O-MOE
`
`CH2CH2OCH3
`
`H
`
`O
`
`H
`
`O
`
`O–
`
`2′-Fluoro
`
`Base
`
`H
`
`O
`
`H
`
`O
`
`H
`
`O P B
`
`H3
`
`O
`
`–
`
`O
`
`O
`
`H
`
`O P C
`
`H3
`
`H
`
`Methylphosphonate
`
`O
`
`Phosphodiester
`
`O
`
`FIG. 1. Select chemical modifi cations that can be incorporated into siRNAs. Modifi cations to the phosphodiester backbone
`are shown, including phosphorothioate, boranophosphate, and methylphosphonate. Modifi cation to the sugar backbone are
`shown, including 2′-fl uoro, LNA, 2′OMe, FANA, and 2′MOE.
`
`The 2′OMe modifi cation is a naturally occurring RNA
`variant and its use in synthetic siRNAs is not anticipated
`to present signifi cant toxicity. Other 2′-modifi cations dis-
`cussed here are not naturally occurring and their potential
`for toxic side effects needs to be considered. The 2′-F modi-
`fi cation has been studied for safety as a component of syn-
`thetic oligonucleotides. While substantially stabilized to
`nuclease attack, even heavily modifi ed duplexes are eventu-
`ally degraded in serum or in cells and processive 3′-exonu-
`clease attack appears to be the primary route of degradation
`(Zou et al., 2008); thus it is expected that free 2′-modifi ed
`nucleosides will be released into cells. When administered
`to rodents, 2′-F residues can later be found incorporated
`into both endogenous DNA and RNA in a variety of tissues
`(Richardson et al., 1999; Richardson et al., 2002). However, no
`
`obvious toxicity was observed to result from exposure to 2’-F
`modifi ed nucleic acids. In contrast, some dose-dependent
`hepatic toxicity has been reported when LNA-containing
`oligonucleotides were administered to mice, ranging from
`mild elevation of hepatic serum transaminases (Fluiter et
`al., 2003) to frank hepatic necrosis (Swayze et al., 2007). In
`another study, no signifi cant toxicity was observed in intra-
`venous (i.v.) dosing of LNA-modifi ed oligonucleotides to
`African green monkeys (Elmen et al., 2008). More studies
`need to be done to fully assess potential toxicity of the LNA
`modifi cation, perhaps considering the possibility of species-
`specifi c variation in such responses.
`Although not generally considered to be a nuclease-stabi-
`lizing modifi cation, substitution of DNA bases into a siRNA
`can alter its stability to ribonucleases and may offer certain
`
`Downloaded by REPRINTS DESK INC from www.liebertpub.com at 03/29/18. For personal use only.
`
`

`

`CHEMICAL MODIFICATION OF SIRNAS FOR IN VIVO USE
`
`309
`
`advantages. A surprisingly large amount of DNA can be in-
`corporated into a dsRNA and retain potency to trigger RNAi;
`the entire S (passenger) strand can be DNA (Hogrefe et al.,
`2006; Pirollo et al., 2007). The guide strand is less tolerant to
`DNA modifi cation than the passenger strand. Nevertheless,
`the 5′ end of the guide strand (AS-strand) can contain DNA
`residues and still retain most functionality (Ui-Tei et al.,
`2008). The ability to insert DNA bases in this area may have
`added benefi t to alter OTEs (see below).
`The modifi cation strategies discussed above are intended
`to impart nuclease resistance to 21-mer siRNAs while retain-
`ing the ability of the duplexes to enter RISC and maintain
`guide-strand function with Ago2. The situation is somewhat
`more complex when introducing chemical modifi cations
`into longer Dicer-substrate siRNAs where some degree of
`nuclease sensitivity is desired so that Dicer cleavage can
`still occur. Thus slightly different modifi cation rules apply
`for this class of siRNAs. So long as the “Dicing domain”
`remains unmodifi ed, very similar patterns of modifi cation
`can be employed with good results (Collingwood et al., 2008;
`Nishina et al., 2008), and it is possible to make duplexes hav-
`ing improved stability in serum while retaining the ability
`to be cleaved by Dicer.
`The relative impact that nuclease stabilization of a siRNA
`has on functional gene knockdown in vivo will vary with
`the method of administration. “Naked” siRNA exposed
`to serum will obviously benefi t the greatest from chemi-
`cal modifi cation. In one study, nuclease-resistant siRNA
`administered by direct hydrodynamic injection gave 4-fold
`greater reduction in target levels than unmodifi ed siRNAs
`(Bartlett and Davis, 2007), however the kinetics of recovery
`after administration were similar. An even greater differ-
`ence in potency between modifi ed and unmodifi ed duplexes
`was seen in a murine hepatitis B virus (HBV) model system
`(Morrissey et al., 2005a). Once the guide strand is associ-
`ated with Ago2 in RISC, the protein components may them-
`selves protect the RNA from intracellular nucleases, and
`unmodifi ed siRNA can show 2–4 weeks duration of action
`in slowly dividing or nondividing cells (Song et al., 2003;
`Omi et al., 2004; Bartlett and Davis, 2007). Thus the great-
`est benefi t from nuclease stability may be realized during
`delivery, and chemical stabilization of the siRNA might be
`less important when employing a delivery tool that protects
`the siRNA cargo.
`
`Off-Target Effect: Triggering the
`Innate Immune System
`
`The innate immune system provides the body’s fi rst line
`of defense in the recognition and response to foreign sub-
`stances. A variety of different specialized receptors have
`evolved that recognize “pathogen associated molecular pat-
`terns” and trigger immune responses when their ligands are
`present (Armant and Fenton, 2002; Tosi, 2005). Some of these
`receptors are restricted to immune cell lineages while others
`are present in all cells. Synthetic nucleic acids can be rec-
`ognized as “foreign” by the innate immune system. These
`responses constitute one of the major types of OTEs which
`must be addressed when using antisense oligonucleotides,
`ribozymes, aptamers, or siRNAs in vivo.
`
`Different receptors exist in the cytoplasm and/or in endo-
`somal compartments that specifi cally recognize foreign
`motifs in DNA or RNA molecules (Marques and Williams,
`2005; Schlee et al., 2006). Toll-like receptor 9 (TLR9) recog-
`nizes unmethylated CpG dinucleotides in DNA. In mam-
`mals, the cytosine in CpG pairs is methylated at the C5
`position; in bacteria, cytosine is not methylated. Synthetic
`DNA made with unmethylated CpG motifs is recognized as
`foreign by TLR9 and triggers a type I interferon (IFN) re-
`sponse (Krieg et al., 1999; Krieg et al., 2000). Similarly, syn-
`thetic RNA can trigger an IFN response. TLR3 recognizes
`motifs in dsRNAs (Alexopoulou et al., 2001) and TLRs 7 and
`8 recognize motifs in ssRNAs (Diebold et al., 2004). In the cy-
`toplasm, oligoadenylate synthetase (OAS), dsRNA respon-
`sive kinase (PKR), retinoic acid inducible gene 1 (RIG-I), and
`melanoma differentiation-associated gene 5 (Mda5) are all
`capable of recognizing dsRNA in some form and can trigger
`immune responses (Stark et al., 1998; Yoneyama et al., 2004).
`TLRs 3, 7, and 8 are largely restricted to endosomal com-
`partments and exposure to these receptors is infl uenced by
`the mode of entry by which a synthetic RNA enters the cell.
`Delivery of siRNA using cationic lipids or liposomes maxi-
`mizes exposure to the endosomal compartment and there-
`fore increases the risk of triggering a TLR based response
`(Sioud and Sorensen, 2003; Ma et al., 2005; Morrissey et al.,
`2005b; Sioud, 2005). A known immunogenic sequence that
`is synthesized within a cell is less immunostimulatory than
`that same sequence transfected into the cell using lipid-
`based delivery (Robbins et al., 2006). Similarly, mechani-
`cal delivery of an immunostimulatory RNA in mice using
`hydrodynamic injection fails to produce an IFN response
`although that same sequence will trigger a response when
`delivered using lipid vehicles (Heidel et al., 2004). Thus the
`route and means of in vivo delivery of a siRNA can play a
`large role in the need to protect that sequence from detection
`from the innate immune system.
`Certain RNA sequences have been identifi ed that are
`particularly immunostimulatory, such as “UGUGU” (Heil et
`al., 2004; Judge et al., 2005) and “GUCCUUCAA” (Hornung
`et al., 2005). Unfortunately, avoiding these sequence motifs
`does not prevent immune activation. The presence of spe-
`cifi c chemical modifi cations allows mammalian tRNAs and
`rRNAs to escape triggering an autoimmune response; pro-
`tective modifi cations include pseudouridine, N6-methyl-A,
`and 2′OMe modifi ed ribose (Kariko et al., 2005). Use of these
`same chemical modifi cations can similarly allow synthetic
`siRNAs to evade immune detection. In fact, inclusion of only
`two or three 2′OMe modifi ed residues in an RNA duplex can
`be suffi cient to prevent immune activation, and modifi cation
`of rU and rG residues is most effective (Judge et al., 2006).
`Mechanistically, 2′OMe groups act as a competitive inhib-
`itor of TLR7 (Robbins et al., 2007). As a result, 2′-modifi ed
`groups can protect in trans, so that modifi cation of only a
`single strand of a siRNA is necessary to block the immuno-
`stimulatory potential of that duplex. Other 2′-modifi cations
`such as 2′-F can confer protection from immune activation; it
`has not been reported if the LNA modifi cation offers similar
`benefi ts. Interestingly, 2′-deoxy (DNA) bases have recently
`been reported to also block immune detection, particularly
`dT or dU bases (Eberle et al., 2008). Triggering TLR7 may
`
`Downloaded by REPRINTS DESK INC from www.liebertpub.com at 03/29/18. For personal use only.
`
`

`

`310
`
`BEHLKE
`
`involve recognition of specifi c structures formed by ssRNAs
`involving internal hairpins, especially containing rU bases
`(Sioud, 2006; Gantier et al., 2008). Given the minimal impact
`that substitution of a limited number of 2′OMe-U or dT resi-
`dues for rU has on functional potency of an siRNA, it may
`be prudent to routinely include either or both of these mod-
`ifi cations in sequences intended for in vivo use, especially
`if lipid-based delivery tools are employed. Unfortunately,
`these modifi cation strategies do not effectively block TLR3
`activation, and immune stimulation through this receptor
`can still occur even in the absence of lipid-based delivery. It
`was recently reported that both modifi ed and unmodifi ed
`naked siRNAs administered by direct intraocular injection
`can trigger cell surface–localized TLR3 and affect angiogen-
`esis pathways in a sequence nonspecifi c fashion (Kleinman
`et al., 2008).
`Defending against the cytoplasmic RNA receptors
`OAS, PKR, RIG-I, and MDA5 may also be important.
`Oligoadenylate synthetase is not fully stimulated until it
`encounters dsRNAs in the 60–70 base size (Minks et al.,
`1979), so direct involvement of OAS in sensing siRNAs seems
`unlikely. Although PKR is generally thought to maximally
`respond to dsRNAs >30 bases long, it can sense the pres-
`ence of species as short as 16 bases. Like TLR7, however, PKR
`recognition can be blocked by a variety of chemical modifi -
`cations (Nallagatla and Bevilacqua, 2008). Interestingly, the
`2′-F modifi cation does not help evade PKR responses. The
`specifi c structural requirements for MDA5 to detect small
`synthetic RNAs have not been defi ned.
`RIG-I is an important cytoplasmic receptor that can sense
`the presence of certain classes of synthetic RNAs and plays
`several different roles in immune surveillance (Yoneyama et
`al., 2004; Kato et al., 2006; Saito et al., 2008). It was reported that
`in certain cell lines synthetic RNA duplexes having 19-mer
`duplex domains with 2-base 3′-overhangs (Dicer-product
`21-mer siRNAs) do not trigger any adverse response but that
`longer RNA duplexes can induce immune responses; TLR3
`dependence was proposed (Reynolds et al., 2006). Although
`some element of TLR3 recognition of these sequences may
`occur, it appears that RIG-I is an important cytoplasmic
`receptor capable of recognizing these longer RNA species.
`Specifi cally, RIG-I can recognize RNA duplexes >21-mer
`length having blunt ends and that the presence of 3′-over-
`hangs evades recognition even for duplexes of 27-mer length
`(Marques et al., 2006). Importantly, 2′OMe modifi cation of
`the RNA duplex will prevent this response and, interest-
`ingly, it seems that the related 2′-F modifi cation is less pro-
`tective (Collingwood et al., 2008; Zamanian-Daryoush et al.,
`2008). RIG-I also recognizes the presence of a triphosphate at
`the 5′-end of nucleic acids (Kim et al., 2004; Hornung et al.,
`2006; Pichlmair et al., 2006), causing an immune response
`to transcripts made by viral polymerases (including in vitro
`transcripts).
`
`Off-Target Effect: Unwanted Participation in
`miRNA Pathways
`
`Although it is possible to achieve single base discrimi-
`nation with select siRNAs, single- or even double-base mis-
`matches are often tolerated and can still reduce target levels
`
`by signifi cant amounts (Du et al., 2005; Birmingham et al.,
`2006; Schwarz et al., 2006; Dahlgren et al., 2008). If the full
`sequence of the reference genome is known, a thorough ho-
`mology screen of candidate siRNAs should permit exclusion
`of sites where unwanted homology exists with other genes
`and theoretically lead to high specifi city. Unfortunately, this
`kind of traditional cross-hybridization analysis can be less
`effective than expected and even carefully screened siR-
`NAs can cause signifi cant changes in expression levels in
`unrelated genes (Jackson et al., 2003; Persengiev et al., 2004;
`Scacheri et al., 2004). It appears that many of these effects
`are mediated by the unintended participation of siRNAs in
`miRNA pathways.
`The miRNA translational suppression pathway is di-
`rected by imperfect base pairing between target and guide
`strand (He and Hannon, 2004; Kim, 2005) and the specifi city
`of this process is defi ned by a 6–7 base “seed region” at the
`5′ end of the antisense strand of the miRNA (Doench and
`Sharp, 2004; Lin et al., 2005; Jackson et al., 2006b). Given the
`expected frequency of fi nding 6–7 base matches between a
`siRNA and nontargeted genes within the entire