RNA interference in biology and disease

RNA interference (RNAi) was first characterized in the nematode worm Caenorhabditis elegans by Fire and colleagues,¹  who found that double-stranded RNA (dsRNA) induced a more potent sequence-specific silencing response than single-stranded antisense RNA alone, which was customarily used for this purpose. Further investigation into this phenomenon demonstrated that injection of dsRNA into the gut of the worm caused a systemic silencing of the target gene that was passed on to the next-generation offspring. The effects of the interference were mediated by a small number of dsRNA molecules per cell, unlike the stochiometric antisense RNA-mediated interference,¹  arguing for a catalytic or amplification element in the interference process. While some model systems, such as plants, share these heritable and systemic interference properties with Caenorhabditis elegans, others such as Drosophila and the adapted mammalian system display transient and cell-autonomous sequence-specific target gene suppression when challenged with dsRNA.

RNAi was first described in plants as an immune response to viral infection. As early as 1928, it was noticed that as tobacco plants infected with tobacco ringspot virus grew, the upper leaves showed resistance to the effects of the virus.²  It is now known that dsRNA intermediates produced during virus infection activate the RNAi machinery to silence expression of complementary genes, thus producing immunity to the virus.^3,4  This defense against foreign genetic material is one of several physiologic pathways that are induced by naturally occurring dsRNAs in a wide variety of eukaryotic organisms including fungi, plants, and animals. With some variations, these responses are all mediated by a common RNAi pathway that involves processing of the dsRNA into short duplexes of about 22 base pairs with characteristic end structure. The RNAi machinery can also be triggered by experimentally introduced synthetic dsRNAs, providing a valuable tool for in vivo gene silencing as described under “RNA interference in mammalian cells.”

In addition to cytoplasmic silencing of gene expression activated by exogenous (viral) dsRNA, RNAi is also responsible for the silencing of cellular mRNAs by endogenous microRNAs (miRNAs).^5,6  miRNAs are small noncoding RNAs that contain inverted repeat regions of complementarity. These repeats lead to the formation of double-stranded hairpins that trigger the RNAi machinery. In plants, miRNAs primarily function by cleaving homologous mRNAs. In animals, however, miRNAs appear to regulate gene expression by targeting partially complementary sequences in the 3′ untranslated region (UTR), which results in translational repression. A large number of miRNAs have been identified. For example, cDNA cloning and computational predictions indicate that the Drosophila and C elegans genomes each contain about 100 miRNAs and vertebrate genomes have about 250. Whereas a large number of miRNAs are known to exist, their target mRNAs and consequent functional importance have only been determined for a few. Nevertheless, based on these findings, primarily in Drosophila and C elegans, it appears that the physiologic role of miRNAs is to guide expression of endogenous genes involved in such processes as developmental timing, cell fate, and cell death. In hematopoietic cells miRNAs have been identified that regulate lineage differentiation.⁷  They may also play a role in the etiology of B-cell chronic lymphocytic leukemias.⁸ 

A third type of gene silencing in response to dsRNA involves the modification of cellular DNA and histones and their packaging into condensed and transcriptionally silent heterochromatin.⁹  The trigger for this type of response appears to be repeat-associated short interfering (si) RNAs generated by hybridization of transcripts from repetitive sequences such as transgene arrays or transposons. Thus, this type of RNAi may also represent a cellular defense mechanism to protect against the potentially deleterious effects of foreign genetic elements such as transposons.

The discovery of dsRNA-induced gene silencing in C elegans allowed genetic screens to be performed that led to the identification of genes required for RNAi in the nematode.¹⁰  Comparison of these genes to those from other species involved in silencing phenomena known as posttranscriptional gene silencing, cosuppression, quelling, and RNAi revealed that all of these events follow a similar core pathway. Nevertheless, there are features of the pathway that show species-specific differences as well as differences depending on the source of the dsRNA trigger. dsRNA molecules, whether introduced experimentally or present as naturally occurring viral byproducts, endogenous miRNAs, or aberrant transgene transcripts, are recognized and cleaved into 21-23 nucleotide siRNAs by the RNaseIII-like enzyme termed Dicer.¹¹  Dicer homologues can be found in S pombe (but not S cerevisiae), C elegans, Drosophila, plants, and mammals (Figure 1). Different species contain different numbers of Dicer homologues and/or associated proteins containing dsRNA binding domains that function to recognize dsRNAs from different sources. For example, in Drosophila, Dicer-1 processes miRNA precursors and Dicer-2 processes long dsRNAs.¹²  In Arabidopsis thaliana there are 4 Dicer homologues that function together with associated proteins to cleave dsRNAs of different types. To date, only one Dicer gene has been identified in mammals, and interacting proteins regulating Dicer function remain to be identified. dsRNA cleavage by Dicer generates siRNAs that contain a 2-nucleotide 3′ overhang^11,13  and a 5′-phosphorylated terminus,^14,15  both of which are required for activity. Processing by Drosophila Dicer is adenosine triphosphate (ATP) dependent and requires a functional RNA helicase domain.¹⁶  In contrast, it appears that human Dicer may not require ATP.^17,18  Processing of miRNAs requires a preliminary processing step in which the long primary transcript is processed by another RNaseIII-like endonuclease, Drosha, within the nucleus.¹⁹  The miRNA precursor is then exported to the cytoplasm where it is processed further by Dicer.

The small dsRNA products of Dicer cleavage are then incorporated into multi-subunit effector complexes. Depending on the species and the source of the dsRNA, different effector complexes are formed with different end results on the target RNA. rasiRNAs are integrated into the RNA-induced transcriptional silencing (RITS) complex and guide chromatin modification. miRNA-containing complexes are usually referred to as miRNPs and direct target RNA degradation (in plants) or translational repression (in animals). Synthetic siRNAs or those derived from naturally occurring long dsRNAs are incorporated into the RNA-induced silencing complex (RISC) and guide its cleavage of target RNAs. RISC has helicase, exonuclease, endonuclease, and homology-searching domains. The initial RISC remains inactive until it is transformed into an active form by the unwinding of the siRNA duplex, through RISC-mediated helicase activity, and loss of the sense strand of the dsRNA molecule.²⁰  This solves the problem of how a stable dsRNA is converted to a form that is capable of using base pairing to search among cellular RNAs for homologous regions. Based on studies aimed at determining the functional characteristics of siRNAs, Zamore and colleagues¹⁴  determined that the antisense strand is mainly responsible for target recognition and silencing activity. The active siRNA/RISC complex targets the mRNA of homologous sequence for degradation and the mRNA is reliably cleaved at regions homologous to the siRNA.²⁰  This degradation is mediated by the endonuclease activity of active RISC.

In order to study gene function in any experimental system, it is useful to eliminate the expression of specific genes and note the resulting effects. In mammalian systems, this has been achieved through the development of knock-out models in mice. While effective, this method has its disadvantages. First, it eliminates gene expression throughout the entire organism, and cannot be used for studying developmental or cell type–specific effects unless tissue-specific knock-outs are employed. Second, generating knock-out models is both labor and time intensive. In theory, these problems could be resolved by adapting RNAi techniques to silence gene expression in vivo in mammalian systems. It was initially thought that this would not be possible since introduction of dsRNA molecules into normal, mature mammalian cells activates an innate antiviral immune response that results in a general inhibition of protein translation and proinflammatory gene expression (Figure 2).²¹  However, in 2001, Elbashir and colleagues²²  and Caplen and colleagues²³  were able to demonstrate RNAi in mammalian systems by the intracellular expression of artificially synthesized mimics of Dicer products, 21-23 base pair siRNA duplexes, which are delivered into cultured cells by transfection. This technique has also been applied to adult mouse models.²⁴  By eliminating long dsRNAs from the process, it was hoped to prevent activation of the innate immune response controlled by the interferon system and its regulator, the dsRNA-activated protein kinase PKR.²¹ 

Introduction of 21 bp siRNAs has allowed for the successful application of RNAi technology to mammalian systems. However, assays using this method are transient in nature and the suppressed phenotype can be lost within several doubling times, most likely due to the dilution of the siRNA. While this approach is reliable for short-term studies of gene expression, it cannot replace knock-out mouse models or allow for precise loss-of-function genetic screens. This is not a concern in organisms such as C elegans or N crassa because they direct ongoing synthesis of siRNAs through the action of an RNA-dependent RNA polymerase. In these systems, the suppressed phenotype is not only maintained, but is also passed on to future generations,²⁵  although the effect gradually diminishes. An additional disadvantage of transient siRNA studies in mammalian systems is that synthesis of the required siRNAs is costly, limiting the benefits from this technique. To address these issues, a system for the stable expression of siRNAs has been developed. Taking clues from the structure of endogenous miRNAs, mammalian expression vectors were designed to direct the intracellular synthesis of siRNAs.^26,27  In most cases, the target-specific insert is made up of a 19-nucleotide sequence complementary to the target, followed by a short spacer and the reverse complement of the same target sequence. Once transcribed, a 19 bp stem-loop structure, termed short-hairpin RNA (shRNA), is processed by Dicer into an siRNA that can direct the down-regulation of target gene expression via the elements of RNAi machinery (Figure 1). Polymerase III promoters, such as T7 or U6, were initially used in these constructs, as they produce siRNAs that mimic the requirements for an efficient siRNA. These requirements include, but are not limited to, the absence of a poly A tail and a termination signal that yields a transcript with a 3′ overhang. Polymerase II–driven shRNA expression vectors have also been developed, which will allow for the regulated expression of siRNAs.

These RNAi vectors include a selectable marker to allow selection of the transfected cells and can also include inducible elements to allow for the regulation of siRNA expression. In cases where transfection efficiencies are low, viral vectors have been designed to deliver the shRNA expression inserts. Lentivirus-based vectors are very effective in infecting noncycling cells, stem cells, and zygotes²⁸  and may become the basis for gene therapy approaches based on RNAi techniques,^29,30  although problems of instability of long-term expression can be encountered.³¹ 

The idea of enzymatically synthesized siRNAs has been applied to kits that are available through Ambion (Austin, TX), which enzymatically synthesize large amounts of siRNAs in vitro using T7 RNA polymerase. While siRNAs produced by this method will also produce a transient phenotype, it allows for the rapid, and relatively inexpensive, production of large siRNA quantities although these can induce nonspecific effects (see “Interferon induction by siRNAs”).

For many years, homology-dependent, RNA-mediated gene silencing techniques have been used as a basic research tool, introducing RNA into cells to interfere with the function of an endogenous homologous gene. These studies assumed that the observed effects resulted from simple antisense mechanisms that depend on Watson-Crick base pairing between the introduced RNA and the endogenous messenger RNA transcripts.³²  However, it is now apparent that the presence of dsRNA accounted for many of the observed robust silencing effects.

In 1998, Fire and colleagues¹  determined that dsRNA was a more efficient inhibitor of sequence-specific suppression of gene expression in C elegans, as opposed to previously employed antisense methods. RNAi can be induced in the nematode worm by direct injection of the dsRNA, by feeding the worm bacteria that have been generated to express siRNAs, or simply by soaking the worm in siRNA-containing media. From that point forward, the emphasis has been not only on understanding how this phenomenon occurs, but also how it can be harnessed as a research tool.³³  Through these investigations, RNAi has been shown to greatly facilitate both “reverse genetic” experiments (identifying the function of a known gene) and “forward genetic” experiments (identifying the gene responsible for a given phenotype). From an application point of view, RNAi may also be useful as a therapy for diseases arising from aberrant gene expression.

Typical reverse genetic experiments involve designing siRNAs (chemically or enzymatically synthesized) or shRNA-expressing constructs targeting a gene of interest. Following transient transfection of siRNAs or selection of shRNA-expressing stable transfectants, the phenotype of the cells is assessed using appropriate functional assays. As discussed below, it is critical in this type of experiment to use adequate controls to ensure that the observed phenotype is due solely to targeting of the gene of interest. Both transient and long-term silencing of the expression of a given gene have also been used in in vivo studies, primarily in mice (reviewed in Paroo and Corey³⁴ ).

Analyses testing gene function through RNAi have been performed in a large number of experimental systems, including hematologic disorders such as acute myeloid leukemia and chronic myelogenous leukemia. Normally, the acute myeloid leukemia (AML) gene product is part of a transcription factor complex whose activity is required for normal hematopoiesis. Chromosomal translocations in hematopoietic malignancies where the AML1 gene is the most frequent target are common in human leukemias.³⁵  Specifically, the translocation t(8;21) resulting in the AML1/MTG8 (myeloid translocation gene) translocation product accounts for 10% to 15% of all de novo cases of myeloid leukemia, although the exact role of the chimeric protein in the development of leukemia was not completely understood.³⁶  To address this RNAi, technology has been used to specifically silence the expression of the AML1/MTG8 translocation product. Electroporation with AML1/ MTG8-specific siRNAs successfully suppressed the expression of the fusion proteins, without interrupting the expression of wild-type AML1 mRNA in 2 t(8;21)–positive cell lines. In this type of functional analysis, Heidenreich and colleagues³⁶  were able to determine the role for the AML1/MTG8 fusion protein in preventing differentiation of the leukemia cells. Suppression of chimeric protein expression in t(8;21)–positive cell lines increased their susceptibility to growth factors that lead to their ultimate differentiation, determined both by changes in cell shape, the display of surface marker proteins, and the up-regulation of the CAAT/enhancer binding protein (C/EBP) marker gene for differentiation. Mismatched siRNAs were used as the controls in these experiments to conclude that these observations were the direct result of siRNA silencing activity, and not the result of nonspecific, off-target effects, as discussed below, under “Nonspecific and off-target effects of RNA interference.”

Clearly, RNAi techniques will also be helpful in investigating related chromosomal abnormalities associated with myeloid malignancies, such as the t(16;21) translocation reported by Heidenreich and colleagues in a recent issue of Blood.³⁶  Chronic myelogenous leukemia (CML) and variants of acute lymphoblastic leukemia (ALL) that arise from a t(9;22) chromosomal translocation resulting in a constitutively active Bcr-Abl tyrosine kinase are also targets for siRNA approaches (Table 1). Scherr and colleagues³⁸  determined that siRNAs recognizing the breakpoint of the fusion protein are effective at specifically silencing its suppression in both established cell lines and primary CML cells, without interfering with the expression of wild type c-abl or c-bcr.

In addition to these type of experiments aimed at silencing a particular gene of interest, RNAi has been combined with genomics to perform large-scale genetic screens aimed at gene discovery. The completed sequencing of the genomes of a number of organisms provides the basis for development of genome-wide libraries of siRNAs. Small-scale efforts using siRNA libraries to identify genes involved in specific signaling pathways, such as the phosphatidylinositol 3-OH kinase and nuclear factor κB (NFκB) pathways, have provided proof of principle for this application.^39,40  In C elegans, RNAi is being used to test the functions of nearly all of the organism's predicted 19 757 genes.³³ 

RNAi likely developed as an endogenous host defense mechanism directed against viral infections. Evidence for this has been seen in both plants^41,42  and animals, such as C elegans⁴³  and Drosophila.⁴⁴  Once RNAi techniques had been successfully developed to suppress endogenous gene expression in mammalian systems, the next obvious question was whether these strategies could be used for therapeutic purposes. Early results from cell culture and animal models suggest this application has the potential to revolutionize the treatment of such diseases as viral infection and cancer, by expressing siRNAs that specifically target components of the virus or the disease for silencing.

Promising studies have already been performed against multiple viruses, including human immunodeficiency virus, influenza virus, and human papiloma virus, preventing the establishment of productive infection in susceptible cells. In addition to pathogenic viruses, RNAi technology has also been used to target specific cancer genes in melanoma,⁴⁵  pancreatic adenocarcinoma,⁴⁶  and leukemia, as discussed here. These studies further illustrate the potential for an RNAi-based gene therapeutic approach.

The RNAi-mediated functional analysis of the AML1/MTG8 fusion protein that was performed by Heidenreich and colleagues³⁶  succeeded not only in advancing our understanding of the basic biology of myeloid leukemia, but also in opening the door for the potential use of RNAi technology in treating this disease. In the case of Bcr-Abl–positive leukemias, the tyrosine kinase inhibitor imatinib mesylate is currently the primary therapeutic agent for this disease. However, suboptimal therapeutic response or resistance is common, preventing disease control. The success reported by Scherr and colleagues³⁸  in suppressing the expression of the Bcr-Abl fusion protein through siRNA expression has provided a potential alternative avenue for treating this disease. In a recent report, Wohlbold and colleagues⁴⁷  were able to silence the expression of the Bcr-Abl fusion protein in leukemia cell lines, resulting in an increased sensitivity to imatinib and γ radiation. In a similar study, Wilda and colleagues⁴⁸  reported an increase in apoptosis in CML cells transfected with an siRNA targeting Bcr-Abl. Clearly there are opportunities to develop RNAi therapeutic approaches to hematologic malignancies.

The initial limitation of RNAi technology is designing an effective siRNA sequence. Advancements in siRNA delivery (such as the enzymatic synthesis of siRNAs from T7 promoters) have placed constraints on which sequences of the target genes can even be considered for use. Many of these constraints depend on the type of polymerase ultimately used to recognize and amplify the siRNA sequence.⁴⁹  However, even following the recommended rules for siRNA design does not ensure effective silencing of the target gene. The efficacy of siRNA-mediated suppression of gene expression depends on a number of factors, including not only the chosen siRNA sequence but also the structure of the siRNA, and the receptiveness of the cell type to siRNA uptake.^50,51  In addition, the half life of the target message and/or protein needs to be considered in order to achieve optimal silencing.

The use of RNAi for therapeutic purposes will depend on other factors as well. Although siRNAs are relatively stable in cell culture conditions, they require enhanced nuclease and thermodynamic stability when in circulation in vivo. Chemical modifications of siRNAs to enhance this stability are being explored.⁵²  While there are mixed opinions as to which type of modification will be most effective at enhancing stability without compromising target silencing activity, advances are being made toward the goal of making siRNAs suitable for therapeutic purposes.

Two of the most important considerations for developing RNAi as a clinical therapy are first, devising efficient mechanisms for delivery of the siRNA to the target cells in vivo, and second, avoiding nonspecific or off-target effects of the siRNA. Since delivery issues apply to all nucleic acid–based therapies and are perhaps less relevant to hematologic applications they will not be considered further. Nonspecific and off-target effects of siRNAs, on the other hand, present an inherent limitation to their usefulness in both basic research and clinical applications and must be carefully considered. Initial studies reporting the successful application of siRNA in mammalian cells led to an assumption of specificity while failing to take into account the multiple roles that short dsRNAs can have on cell metabolism. With increasing use of RNAi as a research tool, it has became clear that introduction of an siRNA molecule into a cell can have multiple effects beyond those caused by gene-specific silencing. In many cases these side effects may not be apparent unless global gene expression studies are performed that look at the effects of siRNAs beyond the expected inhibition of target gene expression. Two general types of nonspecific effects have been observed. First, siRNAs can activate alternative dsRNA-responsive cellular pathways resulting in up-regulation of a large number of genes typically associated with innate immune pathways, including interferon-stimulated genes (ISGs). This represents an “siRNA-specific,” rather than a “target gene–specific,” mechanism of semiglobal gene regulation in which the sequence of the siRNA is irrelevant. A second type of effect, termed off-target, is observed when specific genes other than that which is targeted show altered expression in response to an siRNA. Such sequence-dependent off-target effects are likely due to loose homology requirements for activation of the RNAi machinery in some contexts, such as with miRNA triggers. Finally, since experimental RNAi uses the same cellular machinery as that required for endogenous miRNA-mediated gene regulation, it is possible that the 2 processes could interfere with each other.

Studies of miRNAs have helped to determine the cause of some of these nonspecific effects. miRNAs make up a group of small nontranslated RNAs that regulate the timing of developmental events in an organism. They were initially called small temporal RNAs (stRNAs) because of their sequential pattern of expression and their roles in temporal regulation.⁵²  Similar to siRNAs, miRNAs are 21 to 25 nucleotides in length, mediate the down-regulation of target genes, and are produced as a result of Dicer activity (Figure 2). miRNA sequences are initially localized in the stem of an imperfectly matched stem-loop structure over 70 nucleotides in length. These pre-miRNAs are exported to the cytoplasm, where they are further processed into 21 to 23 nucleotide mature miRNAs by Dicer.^53,54  miRNAs appear to be the primary products of Dicer activity during unstressed conditions⁵⁵  and down-regulate target gene expression after transcription by binding to a region of partial complementarity in the 3′ UTR of the target mRNAs. Since such similar formation pathways exist for the 2 types of short dsRNAs, it was hypothesized that siRNAs are capable of miRNA activity. Gene expression studies aimed at exploring this possibility determined that a subset of siRNAs can induce the posttranslational suppression of target gene expression, as seen with miRNA activity. In addition, these nonspecific inhibitory effects are seen at low siRNA concentrations and only partial complementarity to the suppressed gene is required. While a single mismatch between an siRNA and its target will reduce specific silencing efficiency, that same siRNA may still be able to down-regulate the expression of nontargeted genes that contain regions of partial complementarity.^56,57  When computational analyses were performed on the siRNAs, multiple potential 3′ UTR targets were identified that contain partial sequence homologies to each strand of a given siRNA.⁵⁷  If siRNAs can indeed act as miRNAs and control the expression of multiple genes, global expression profiles must be analyzed to ensure siRNA specificity, especially when these studies are used to determine gene function.

Interferons (IFNs) are cytokines that function as the first line of defense against viral infection in mammals.⁵⁸  They can be induced in response to molecular patterns specific to pathogens, including bacteria, viruses, and fungi, and are secreted to provide paracrine protection against virus infection and to mediate adaptive immune responses. Activation of this innate immune response is efficiently triggered by dsRNA, a pathogen-associated molecular pattern (PAMP) commonly formed during the replication cycle of most viruses. dsRNA is responsible for the initiation of 3 known types of signaling events involved in the antiviral response. Initially, dsRNA binds to constitutively expressed dsRNA recognition proteins. These proteins act either to directly mediate antiviral events or to initiate signaling cascades that result in the up-regulation of IFNs and other antiviral proteins (Figure 3). dsRNA can also directly activate the transcription of specific genes through alternative, IFN-independent pathways (Figure 3).

IFNs induce the transcriptional activation of interferon-stimulated genes (ISGs), whose protein products confer cellular antiviral or antiproliferative activity. Many protein products of these antiviral genes are constitutively expressed in uninfected cells. These proteins, such as PKR and 2′-5′A synthetase, are not only up-regulated through IFN-mediated transcriptional activation, but are also present at basal levels to directly respond to the initial infection. Following activation by dsRNA, PKR dimerizes, autophosphorylates, and subsequently phosphorylates its substrates, the best characterized of which is the alpha subunit of eukaryotic initiation factor 2 (eIF2α). Phosphorylation of eIF2α by PKR renders the initiation factor unavailable for further rounds of translation. As a result, PKR inhibits both viral replication⁵⁹  and cellular protein synthesis.⁶⁰  PKR also acts as a signal transducer in pathways leading to the activation of NFκB and ATF2.⁶¹ 

A family of 2′-5′ oligoadenylate synthetases (OASs) are also activated by dsRNA. Activated OASs convert ATP into short 2′-5′ linked oligoadenylates termed 2-5A molecules. 2-5A binds and activates RNase L, resulting in the nonspecific degradation of cellular single-stranded RNA, and the onset of apoptosis.^62,63  In addition to these pathways, dsRNA is also able to trigger signaling events mediated by Toll-like receptor 3, as well as the up-regulation of genes directly controlled by dsRNA expression via activation of transcription factor IRF3^64,65  (Figure 3).

Although siRNAs were originally thought to be too small to induce a dsRNA-initiated response inside the cell, up-regulation of certain classic IFN-stimulated genes, such as GBP1, CCL2, FGF2, CXCL11, and ISG20, can be detected.⁶⁶  The observed up-regulation was not seen in the control or mock-transfected samples, indicating that the effect was due to the intracellular presence of the double-stranded siRNA. These nonspecific effects were concentration dependent and appeared to be consistent for all chemically synthesized siRNAs tested. Microarray experiments revealed that chemically synthesized siRNAs trigger the activation of signaling components that overlap, but are not identical to, those regulated by IFN.⁶⁷  Similarly, the up-regulation of the ISG Stat 1, in addition to activation of PKR, has been seen in response to multiple chemically synthesized siRNAs.⁶⁷  Thus, it is clear that at least in some circumstances PKR activation can occur in response to dsRNAs smaller than 30 base pairs.

While recent optimization studies of siRNAs through efficacy-determining algorithms eliminate many of these effects, significant ISG up-regulation by siRNAs has been noted. Not all chemically synthesized siRNAs cause ISG up-regulation, as determined in most cases by observing the resulting levels of OAS2, a highly inducible ISG.^57,68  There are a number of possibilities to explain this inconsistent trend, including siRNA sequence specificity and/or cell type specificity. However, it is important to note that ISGs reflect secondary effects subsequent to IFN induction. It is also necessary to measure primary gene activation events that can occur in the absence of IFN induction. Many cultured tumor cell lines have defects in dsRNA and or IFN signaling and restricting measurements to these pathways may miss other nonspecific events. In addition, not all siRNAs targeting a particular gene are effective to the same degree at silencing expression. Because no obvious trends can be determined that govern the level of efficiency of a given siRNA, such as secondary structure of the target sequence site or location of the target region within the gene, a systematic analysis is often needed to define specific determinants of siRNA efficiency. Different studies have set out to determine the siRNA characteristics that increase functionality while minimizing the nonspecific effects. These 2 ideals may not be mutually exclusive; a highly efficient siRNA can be used at lower concentrations, reducing the potential for siRNA concentration-dependent, nonspecific side effects. In one study in particular, 8 characteristics were identified to be associated with siRNA functionality. The application of an algorithm that incorporates these criteria in siRNA design greatly improves the success rate of siRNA selection.⁶⁹  Under the specified conditions, RNAi assays using chemically synthesized siRNAs have been optimized and nonspecific side effects as a result of this technique have been minimized.

Nonspecific effects can occur to a much higher degree in response to siRNAs synthesized from Pol III–driven promoters. While very effective at silencing target gene expression, siRNAs synthesized from Type III RNA polymerases (such as T7 and U6) often result in the production of more robust nonspecific side effects. It was initially noted that siRNAs produced from both the expression of endogenous shRNA-producing vectors⁶⁸  and from the transfection of in vitro transcribed siRNAs⁶⁷  induce a robust interferon response. Microarray studies performed in our laboratory illustrate the concentration dependent, global up-regulation of IFN-stimulated genes in response to T7-synthesized siRNAs.⁶⁷  These nonspecific effects represent a bona fide IFN response, unlike the response seen in early studies using nonoptimized chemically synthesized siRNAs, where only a subset of ISGs are up-regulated, and can result in an antiviral state in siRNA transfected cells (C.A.S. and B.R.G.W., unpublished observations, August 2003).

While the direct cause of these nonspecific effects is still being explored, evidence has been presented that shows that the Pol III promoters, which produce siRNAs that effectively mimic natural siRNA traits, may be one source of the problem. In 2 independent reports by Kim et al⁷⁰  and Pebernard and Iggo,⁷¹  the 5′ region of these enzymatically produced siRNAs contain elements that could be traced as responsible for ISG induction. Regardless of the exact nature of the elements, it does appear that component(s) near the 5′ end of Pol III–driven transcripts contributes to, but do not account for all, ISG induction by these constructs.

Because the benefit of RNAi applications to both basic and applied research is substantial, strategies must be developed to maintain a high level of siRNA efficiency while minimizing the potential for the misinterpretation of data due to the nonspecific or off-target effects that have been associated with siRNA expression. While in some cases the production of a nonspecific IFN-mediated response could be beneficial therapeutically, such as in treating CML, for other indications off-target effects could result in unexpected side effects. This can already be avoided to a large extent by following some basic guidelines of good experimental practice. These include (1) using available information and algorithms to design the most effective and specific siRNA possible so that it can be used at very low concentrations, (2) using several siRNAs against the same target since it is unlikely that they all would have similar sequence-dependent off-target effects, (3) using different control siRNAs against measurable irrelevant genes, and (4) rescuing the phenotype caused by an siRNA by ectopic expression of a version of the gene that cannot be silenced by the siRNA. Finally, as our understanding of this fascinating mechanism of gene regulation improves, new ways to effectively harness RNAi for experimental and therapeutic approaches will be revealed.

We thank Patricia Stanhope-Baker for comments on the manuscript and editorial assistance.