Skip to main content
NCBI home page
Molecular Therapy logoLink to Molecular Therapy
. 2011 Sep 27;19(12):2169–2177. doi: 10.1038/mt.2011.185

Rational Design of Therapeutic siRNAs: Minimizing Off-targeting Potential to Improve the Safety of RNAi Therapy for Huntington's Disease

Ryan L Boudreau 1, Ryan M Spengler 2, Beverly L Davidson 1,3,4
PMCID: PMC3242660  PMID: 21952166

Abstract

RNA interference (RNAi) provides an approach for the treatment of many human diseases. However, the safety of RNAi-based therapies can be hampered by the ability of small inhibitory RNAs (siRNAs) to bind to unintended mRNAs and reduce their expression, an effect known as off-target gene silencing. Off-targeting primarily occurs when the seed region (nucleotides 2–8 of the small RNA) pairs with sequences in 3′-UTRs of unintended mRNAs and directs translational repression and destabilization of those transcripts. To date, most therapeutic RNAi sequences are selected primarily for gene silencing efficacy, and later evaluated for safety. Here, in designing siRNAs to treat Huntington's disease (HD), a dominant neurodegenerative disorder, we prioritized selection of sequences with minimal off-targeting potentials (i.e., those with a scarcity of seed complements within all known human 3′-UTRs). We identified new promising therapeutic candidate sequences which show potent silencing in cell culture and mouse brain. Furthermore, we present microarray data demonstrating that off-targeting is significantly minimized by using siRNAs that contain “safe” seeds, an important strategy to consider during preclinical development of RNAi-based therapeutics.

Introduction

RNA interference (RNAi) directs sequence-specific gene silencing by double-stranded RNA which is processed into functional small inhibitory RNAs (~21 nucleotides).1 In nature, RNAi for gene regulation occurs primarily via small RNAs known as microRNAs (miRNAs).2 Mature miRNAs (~19–25 nucleotides) are processed from larger primary miRNA transcripts which contain stem-loop regions. Via a series of processing events catalyzed by the ribonucleases, Drosha and Dicer,3,4 the miRNA duplex region is liberated, and a single strand (the antisense “guide” strand) is then incorporated into the RNA-induced silencing complex,5,6 thus generating a functional complex capable of base-pairing with and silencing transcripts by various means depending on the degree of complementarity. A high-degree of base-pairing causes target transcript cleavage, whereas imperfect binding (typically to transcript 3′-UTRs) induces the canonical miRNA-based repression mechanism, resulting in translational repression and mRNA destabilization.7 Indeed for the latter, pairing via the seed region with as few as 6–7 base pairs may be sufficient to trigger silencing.8

Elucidating the mechanisms involved in endogenous miRNA biogenesis and gene silencing has enabled scientists to devise strategies to co-opt the cellular RNAi machinery and direct silencing of virtually any gene of interest using siRNAs, short-hairpin RNAs (shRNAs), and artificial miRNAs;9 the latter two represent expressed stem-loop transcripts which release siRNAs upon processing.10,11 siRNAs are generally designed with the guide strand exhibiting perfect complementarity to the intended mRNA target to promote cleavage. This potent gene silencing approach has become a powerful molecular tool to study gene function and is being developed as a therapeutic strategy to suppress disease-causing genes. The utility of siRNA-based technologies as biological or clinical interventions is largely limited by our abilities to design effective and specific inhibitory RNAs and to introduce them into target cells or tissues. A major consideration for gene silencing applications is specificity, and there is mounting evidence supporting that siRNAs bind to and repress unintended mRNAs, an effect known as off-target silencing.12,13,14 Off-targeting primarily occurs when the seed region pairs with 3′-UTR sequences in mRNAs and directs translational repression and destabilization of those transcripts.15 Recent data supports that seed-based off-targeting may induce toxic phenotypes.16 Anderson et al. observed that the magnitude of siRNA off-targeting is directly related to the frequency of seed complements (hexamers) present in the 3′-UTRome.17 By evaluating subsets of siRNAs with differing off-targeting potentials (low, medium and high; predicted based on hexamer distributions in human 3′-UTRs), they discovered that siRNAs in the low subset had significantly diminished off-target signatures (based on microarray data) and less adverse effects on cell viability as compared to siRNAs in the medium and high subsets. These observations established the importance of considering seed complement hexamer distributions as a key criterion for designing highly specific siRNAs, and some siRNA design tools have since incorporated seed-specificity guidelines into their algorithms.18,19,20 However, most publically available algorithms remain strongly biased for gene silencing efficacy over specificity, and thus, very few candidate siRNAs actually contain seeds with low off-targeting potentials. This is revealed in a literature survey of siRNAs under therapeutic development; only 7 of 80 recently published siRNAs with therapeutic relevance (Supplementary Table S1) could be classified into the low off-targeting subgroup as per the Anderson et al. study. This may be problematic as siRNAs move into early stage clinical trials. While potency-based design is rational, current publicly available tools identify only a fraction of the functional siRNAs for a given target transcript, and often times, highly functional siRNAs do not satisfy several design rules. For these reasons, and in the interest of improving the safety profile of therapeutic RNAi, we hypothesized that a siRNA design scheme prioritizing specificity yet promoting efficacy would yield candidate siRNA sequences with minimal off-targeting potential and a robust capacity for potent gene silencing.

Results

Some artificial miRNAs induce sequence-specific toxicity

Previous studies from our laboratory and others' have demonstrated the potential of RNAi therapeutics for treating Huntington's disease (HD), a dominant neurodegenerative disease caused by a CAG repeat expansion which confers a toxic gain-of-function to the resulting huntingtin (htt) protein.21 In several rodent models for HD, viral-based expression of RNAi hairpins targeting mutant htt mRNA in brain reduced transcript and protein levels by ~50–70%, improving behavioral and neuropathological phenotypes.22,23,24,25,26 Following these proof-of-concept successes, we initiated studies to evaluate and optimize the safety of RNAi-based therapeutics. We compared the silencing efficacy and safety of shRNA and artificial miRNA expression vectors in vitro and in vivo. We found that shRNAs are more potent but induce toxicity in cell cultures and in mouse brain, whereas artificial miRNAs are expressed at tolerably lower levels and display better safety profiles while maintaining potent gene silencing.27,28,29 Since this discovery, we have tested several artificial miRNA sequences in mouse brain using recombinant adeno-associated viruses (AAV serotype 2/1) for delivery, and in some instances, have observed sequence-dependent toxicity. For example, one artificial miRNA targeting htt (miHD-Ex1) caused a high incidence of seizures and morbidity in treated mice (data not shown); of note, this toxic phenotype was not a consequence of htt knockdown, as we and others have previously reported that silencing endogenous htt in mouse brain is tolerated.25,26 In another instance, a nontargeted artificial miRNA (miSCR, a scrambled control) induced evident neurotoxicity as indicated by increased staining for Iba1, a marker for resting and reactive microgila, in treated regions of the striatum (Figure 1). Together, these data suggest that although artificial miRNAs show improved safety over shRNAs, sequence-dependent toxicity remains a concern. We therefore explored supplemental means to improve safety by employing a rational siRNA design scheme intended to minimize the probability for off-target silencing.

Figure 1.

Figure 1

Open in a new tab

The artificial microRNA (miRNA), miSCR, causes neurotoxicity in mouse brain. Wild-type mice were injected into the striatum with adeno-associated virus (AAV)-green fluorescent protein (GFP) (expresses GFP only) or AAV-miSCR-GFP (expresses both the artificial miRNA and GFP), and histological analyses were performed on brains harvested at 6 months post-treatment. Photomicrographs representing GFP autofluorescence and immunohistochemical staining of IbaI-positive microglia are shown. Bars = 200 and 50 µm for ×10 and ×40 images, respectively. Boxed inset represents region of ×40 image.

Selection and screening of htt-targeting siRNAs with low off-targeting potentials

The aforementioned Anderson et al. study correlated siRNA toxicity potentials with seed complement frequencies in the human 3′-UTRome.17 Here, we estimated the number of potential off-target transcripts for each hexamer by determining the number of human RefSeq 3′-UTRs containing a specified hexamer (out of the 4,096 possible) (Supplementary Table S2). Similar to the findings by Anderson et al., the majority of hexamers are present in ~4,000 3′-UTRs or more, and interestingly, there is a distributional peak containing 1,135 hexamers which are present in 2,000 3′-UTRs or less (Figure 2). Because siRNAs are typically designed to target coding regions, we determined the probability of finding these relatively rare hexamers in human RefSeq coding exons. This was ~14%, suggesting that 1 in 7–8 randomly designed siRNAs would contain these rare hexamers in the seed region. To improve upon this nominal probability, low-frequency hexamers may first be located within target transcript sequence and subsequently used as a foundation for designing siRNAs with minimal off-targeting potentials. For example, we scanned the human htt coding sequence for low-frequency hexamers, and with each instance, examined the nearby context to determine whether the siRNA containing the hexamer seed complement would satisfy two criteria: (i) faithful loading of the intended antisense guide strand and (ii) GC-content between 20 and 70% (Figure 3a). Not only do these attributes represent the most prominent determinants of siRNA potency,18 but proper loading of the antisense guide strand is mandated to mitigate potential off-targeting mediated by the sense “passenger” strand. Strand-loading is dictated by the thermodynamic properties present at the siRNA duplex ends,5,6 with guide strand loading encouraged by weak pairing (A/G-U) at the 5′ end and strong G-C binding at the opposing terminus (Figure 3a). Of note, we apply this principle to the terminal two base-pairs at each end and take advantage of weak G-U wobble pairing to impart instability at the 5′ end of the antisense strand when applicable.30 Finally, we select siRNAs based on a fairly liberal range of GC-content (20–70%) which supports a suitable potential for efficient silencing (>80%), as determined by our evaluation of large-scale knockdown data for 2,431 randomly designed siRNAs targeting 31 unique mRNAs (data not shown).31 As with most siRNA design algorithms, candidate siRNA sequences satisfying the above criteria are subjected to BLAST to evaluate the potential for off-target cleavage events mediated by near-perfect complementarity to unintended mRNAs (see ref. 18 for BLAST parameters).

Figure 2.

Figure 2

Open in a new tab

Overview of seed-related off-targeting: mechanism and probabilities. (a) Diagram depicting the expression and processing of an artificial microRNA (miRNA) to produce the mature small inhibitory RNA (siRNA) duplex. The antisense guide strand is loaded into RNA-induced silencing complex (RISC) and may direct on-target silencing (intended) and off-target silencing (unintended). (b) Cartoon highlighting the relationship between the frequencies of seed complement binding sites in the 3′-UTRome and the off-targeting potential for siRNAs. (c) The number of human mRNA 3′-UTRs containing a given hexamer was determined for all of the 4,096 possible hexamers and a binned distribution is shown. The probabilities that randomly selected siRNAs targeting human coding sequence (CDS) will contain seed complements in a given range (white and gray shading) are also presented. For example, there is only a 10% chance that a randomly selected siRNA contains a seed complement for a hexamer present in ~1,500 human 3′-UTRs or less. Note: the sequences tested in this article are placed above their respective ranges.

Figure 3.

Figure 3

Open in a new tab

Selection and screening of huntingtin (htt)-targeting small inhibitory RNAs (siRNAs) with low off-targeting potentials. (a) Schematic outlining the selection of “safe” seed siRNAs with proper strand-biasing. (b) Plasmids expressing artificial microRNAs (miRNAs), harboring the indicated siRNA sequences, were transfected into HEK293 cells, and quantitative PCR (QPCR) analysis was performed 24 hours later to measure endogenous htt mRNA levels. U6 (promoter-only) and HD2.4 (a previously published htt RNAi sequence) serve as the negative and positive controls, respectively. Results are shown as mean ± SEM (N = 6, *P < 0.001, relative to U6).

Using our siRNA design criteria (low off-targeting potential seed, strand-biasing, and GC-content), we initially identified eight htt-targeting candidate sequences for further testing. We embedded the siRNA sequences into the context of our U6-driven artificial miRNA-based expression vectors (Figure 2a) and screened them for gene silencing against endogenous htt in HEK293 cells (Figure 3b). We observed two candidates (HDS1 and HDS2) that effectively silence htt mRNA (~50%, relative to control). Notably, this magnitude of in vitro silencing against endogenous htt is comparable to the levels achieved by other htt RNAi sequences (including HD2.4) that previously showed therapeutic efficacy in HD mouse models.22,23,25

Microarray analyses of seed-related off-targeting

To validate the low off-targeting potential of these effective sequences (HDS1 and HDS2) and our siRNA design scheme, we performed microarray analysis to assess seed-related off-target gene silencing. We included several RNAi constructs which target human htt and various control sequences to help discern off-target gene silencing from gene expression changes that result from suppressing htt (Table 1). Of note, all sequences used were designed to promote proper loading of the antisense strand to avoid the confounding potential of off-targeting mediated by the passenger strand. The htt-silencing group consisted of HDS1, HDS2, HD2.4, and HD8.2; the latter two were previously designed without regard for the seed sequence and have >4,500 potential off-target transcripts each28 (Figure 2c). The control group (i.e., non-htt-targeting) consisted of several sequences (8.2mis, 8.2-124a, Terror, and Safe), each designed to serve a unique purpose (Table 1). 8.2mis contains the same seed as HD8.2 but has central mismatches to prevent htt silencing, whereas 8.2-124a and Terror are HD8.2 scrambled sequences which respectively contain a seed mimic of miR-124a (a naturally occurring and highly conserved miRNA) and a seed with high off-targeting potential (i.e., complements a highly abundant hexamer in the human 3′-UTRome). Of note, 8.2-124a was included as a control for detecting seed-related off-targeting within the microarray data and to underscore the prospective concern of designing siRNAs (scrambled controls or on-target sequences) such that they unintentionally contain naturally occurring miRNA seeds. Finally, the Safe construct contains an arbitrary sequence designed to have low off-targeting potential based on 3′-UTR hexamer frequencies. Together, these constructs provide a wide-range of off-targeting potentials and address problems that can inadvertently arise when including scrambled sequences as controls in RNAi experiments, a commonly used practice.

Table 1. Microarray constructs.

graphic file with name mt2011185t1.jpg

Open in a new tab

We carried out transcriptional profiling in cultured HEK293 cells 72 hours after transfection with RNAi expression plasmids (N = 4 per construct). Initially, gene expression changes were detected by performing ANOVA statistical analysis using all treatments included in the study. As anticipated, htt was consistently among the most significantly downregulated transcripts in samples treated with htt-targeting RNAi sequences (P < 5 × 1011, relative to U6), and these microarray data were corroborated by quantitative PCR (QPCR) evaluation of htt mRNA levels in the same RNA samples (Figure 4a). Next, we performed hierarchical clustering using differentially expressed genes within the dataset (P < 0.0001, 827 genes) to measure the relatedness among the various treatments. These include gene expression changes which occur as a result of knocking down endogenous htt in addition to sequence-specific off-targeting events. Notably, we observed a closer relationship between the low off-targeting potential sequences (Safe, HDS1, and HDS2) and the U6 promoter-only control as compared to the remaining sequences, which were designed either blindly (HD2.4 and HD8.2) or intentionally with mid-to-high off-targeting potentials (8.2-124a and Terror). These clustering results support a clear association between off-targeting potential and impact on the transcriptional profile (Figure 4b), corroborating the Anderson et al. observations.17 In addition, these data substantiate that changes related to off-targeting are more robust than those resulting from htt silencing. Visualization of the complementing heatmap made obvious the overwhelming amount of off-targeting caused by Terror and, to a slightly lesser degree, 8.2-124a (Figure 4b). The overlap between these sequences is likely due to their seed similarity (Table 1), and subsequent analyses confirmed that much of this off-targeting was seed-related. For the sequences with low-to-mid off-targeting potentials, the relationship between off-targeting potentials and gene expression profiles was better visualized by removing the Terror and 8.2-124a samples from the ANOVA analysis and repeating hierarchical clustering of differentially expressed genes (P < 0.01, 985 genes) (Figure 4c). With this approach, the heat maps showed gene suppression signatures that were unique to each of these sequences, with the exception of HD8.2 and 8.2mis. As previously noted, these constructs share the same seed sequence, and this evident overlap affirms that much of the observed gene expression changes are seed-related, rather than caused by htt knockdown. In addition, this example highlights the benefit of designing on-target and control siRNA sequences that share the same seed. This preserves off-targeting between the two sequences and is therefore beneficial when applying RNAi-based tools to study gene function or validate drug targets.

Figure 4.

Figure 4

Open in a new tab

Evaluation of microarray data for huntingtin (htt) silencing and off-targeting. HEK293 cells were transfected with U6 promoter-only or U6-driven artificial microRNA (miRNA) expression plasmids (n = 4 for each treatment), and RNA was harvested 72 hours later for microarray analysis. Two-way ANOVA was performed to detect differentially expressed genes among the treatment groups. (a) Htt mRNA levels determined by microarray (gray bars) were consistent with those measured by quantitative PCR (QPCR) (black bars) using the same RNA samples. (b) Hierarchical clustering and heat-maps were generated using differentially expressed genes (P < 0.0001, 827 genes) to visualize the relationships among the treatment groups. Interestingly, all of the “safe” seed sequences are more related to U6 than the remaining sequences predicted to have higher off-targeting potentials (boundary marked by white line). (c) Hierarchical clustering and heat-maps were generated using differentially expressed genes (P < 0.01, 985 genes) to visualize the relationships among the treatment groups. The impact of seed sequence on gene expression can be appreciated by the clustering of HD8.2 and 8.2mis which share the same seed. Notably, the predicted low off-targeting sequences (Safe, HDS1 and HDS2) are more similar to U6 and have smaller off-targeting signatures compared to both HD2.4 and HD8.2. Seed-related off-targeting was evaluated by (d) cumulative distribution and (e) motif discovery analyses. (d) Cumulative distribution plots for gene expression values are shown for transcripts containing (1 site or 2+ sites) or lacking (baseline) 3′-UTR seed complement binding sites for the indicated sequence and strand. A shift to the left indicates an increased likelihood of being downregulated. AS, antisense, S, sense. Kolmogorov–Smirnov (KS)-test P values are shown; N.S., no statistical significance (P > 0.1). (e) Motif discovery analyses identified an enrichment of seed complement binding sites in the 3′-UTRs of downregulated genes (>1.1-fold) unique to each treatment. Shown here are the examples of 8.2-124a and Terror; similar data for the remaining sequences supports that each mediates detectable seed-related off-targeting to some degree (Supplementary Figure S1).

We next assessed whether the observed gene expression changes could be explained by seed-mediated gene silencing. Cumulative distribution analyses of gene expression levels indicated that transcripts containing seed binding sites for the antisense strand in their 3′-UTR had a much higher probability of being downregulated (i.e., curve shifting left) (Figure 4d, top and Supplementary Figure S1a), and the degree of downregulation was dependent upon the number of binding sites present, consistent with previous reports characterizing miRNA seed-mediated silencing of target transcripts.32 We also performed cumulative fraction analyses to detect seed-related gene silencing caused by the passenger strand; in this case, the presence of 3′-UTR-binding sites had little to no detectable influence on gene expression, supporting that our vector design (i.e., two strong G-C base-pairs at the sense 5′ and two weak A/G-U base-pairs at the sense 3′) promotes proper strand-biasing (Figure 4d, bottom). As a complementary approach to detect seed-related gene silencing events, we performed motif discovery analyses using 3′-UTR sequences of downregulated transcripts unique to each treatment group. In all instances, we found significant enrichment of motifs complementary to the respective seed sequences in the uniquely downregulated transcript 3′-UTRs relative to a background 3′-UTR dataset consisting of all known human 3′-UTRs (Figure 4e and Supplementary Figure S1b). These data confirm that seed-related off-target gene silencing is present in the datasets for all RNAi sequences tested. Upon further evaluation, we estimated the number of seed-related off-targets for each RNAi sequence by identifying transcripts that were downregulated (1.1-fold, P < 0.05, relative to U6) and contain the relevant seed complements in their 3′-UTR (Table 2). This analysis revealed that using our “safe” seed design method, HDS1 and HDS2 show nearly a log improvement in minimizing seed-related off-targeting, as compared to our previous lead candidates, HD2.4 and HD8.2.

Table 2. Off-target summary.

graphic file with name mt2011185t2.jpg

Open in a new tab

In vivo silencing and safety of HDS sequences

Having identified that HDS1 and HDS2 have substantially fewer seed-related off-targets, we next tested these sequences for silencing and safety in vivo in mouse brain. We intrastriatally injected AAV1-miHDS1, AAV1-miHDS2 or AAV1-miSafe (control) into two cohorts of wild-type mice. Of note, HDS1 exhibits full complementarity to mouse, rhesus and human htt sequences, making it an attractive candidate for preclinical testing. HDS2 only targets human htt, with mismatches to the corresponding mouse and rhesus target sequences. At three weeks postinjection, we performed QPCR analyses to evaluate gene silencing efficacy in striatal tissue harvested from the first cohort of animals and observed significant htt mRNA knockdown (~60%) in mice treated with AAV1-miHDS1, relative to uninjected and AAV1-miSafe-treated mice (Figure 5a). Notably, previous reports from our laboratory and others' demonstrate that ~60% silencing of striatal htt transcripts in HD mouse models markedly reduces protein levels, resulting in appreciable therapeutic efficacy.22,25 The second cohort of mice was sacrificed at six months postinjection to evaluate long-term vector tolerability. Staining for Iba1, a marker for resting and reactive microgila, showed no evidence for neurotoxicity in transduced regions of the striata, relative to nearby untransduced tissue (Figure 5b; refer to Figure 1 for comparison to miSCR, a toxic sequence with high off-targeting potential). These results are encouraging considering that HD2.4, previously shown to be therapeutically efficacious in short term studies, caused modest but still detectable increases in Iba1 staining in both wild-type and HD mice.25,28 These in vivo data warrant further preclinical testing of HDS1 and HDS2. Furthermore, the data corroborate previous reports demonstrating that reducing wild-type htt mRNA levels by ~60% in mouse striatum does not induce overt neurotoxicity.25,26

Figure 5.

Figure 5

Open in a new tab

Silencing efficacy and safety of HDS sequences in mouse brain. Wild-type mice were injected into the striatum with adeno-associated virus (AAV) viruses coexpressing artificial microRNAs (miRNAs) and green fluorescent protein (GFP). (a) At 3 weeks postinjection, GFP-positive striata were harvested and quantitative PCR (QPCR) analysis was performed to measure endogenous mouse huntingtin (htt) mRNA levels. Results are shown as mean ± SEM (n ≥ 3, *P = 0.001, relative to uninjected striata). (b) Brains from additional cohorts of injected mice were harvested at 6 months postinjection and histological analyses were performed to assess neurotoxicity. Photomicrographs representing GFP autofluorescence and immunohistochemical staining of IbaI-positive microglia are shown. Bars = 200 and 50 µm for ×10 and ×40 images, respectively. Boxed inset represents region of ×40 image.

Discussion

Although the absolute specificity and safety of RNAi approaches remains questionable, recent advances in understanding RNAi-induced toxicities (e.g., pathway saturation and off-targeting) are facilitating researchers in devising strategies to limit these adverse events. For example, the discovery that high-level shRNA expression causes lethality in mice33 prompted us to test alternative hairpin-based vectors (e.g., artificial miRNAs) for their capacity to limit the production of RNAi substrates following viral-based delivery in vivo, thus resulting in improved tolerability.28,29 Furthermore, Anderson et al. recently evaluated the impact of 3′-UTR seed complement frequencies on siRNA off-targeting potentials, using a set of randomly designed siRNA sequences targeting a variety of genes.17 Low off-targeting potential siRNAs were found to exhibit higher specificity as per mRNA profiling, lower toxicity and fewer false positives in phenotypic screens. The authors proposed that siRNAs with low seed complement frequencies improve the accuracy of RNAi screens to study gene function or validate drug targets. Here, we took advantage of these findings to deliberately design therapeutic siRNAs with low off-targeting potentials, as a means to promote safety in preclinical development of RNAi therapy for HD. We identified two candidates (HDS1 and HDS2) which effectively silence human htt mRNA, induce minimal seed-related off-targeting and are well-tolerated in mouse brain long-term.

Although our work was initially undertaken to develop siRNAs with low off-targeting potentials, a similar strategy may be employed to intentionally design siRNAs with high off-targeting capacities (e.g., Terror sequence) for use as antitumor agents. This approach may deter tumor escape by more broadly disrupting essential cellular pathways, as compared to delivering siRNAs targeting specific oncogenes, a strategy under current investigation.34

Researchers using RNAi triggers in basic and translational research often employ scrambled sequences as controls. Our work highlights the importance of carefully designing control siRNAs, with attention to putative seed sequences that can inadvertently induce considerable off-target silencing via miRNA-based mechanisms. Here, we intentionally introduced either a known miRNA seed (8.2-124a) or a seed with high off-targeting potential (Terror) into scrambled siRNA sequences. As expected, both induced significant seed-related alterations in transcriptional profiles, when compared to our control vector (Safe) designed to exhibit low off-targeting potential. Furthermore, we describe and test the design of a “same seed” control vector (8.2mis). This approach resulted in significant preservation of off-targeting relative to the corresponding on-target sequence (HD8.2). These data encourage the use of “same seed” controls in future RNAi experiments.

There are several key considerations which apply to “safe” seed siRNA design. First, low off-targeting potential does not necessarily mean nontoxic, as off-target identity remains a crucial influence on tolerability. Our improved ability to accurately identify high probability off-targets allows us to better select lead candidate siRNAs, particularly when several low off-targeting sequences are available for a given target sequence. Second, observed safety in preclinical toxicity studies in either rodents or nonhuman primates may not ensure success in humans, as differences in 3′-UTR sequences creates off-targeting profiles unique to each species.35 It is important to note, that although off-target identities may be species-specific, the off-targeting potentials for each hexamer remain highly consistent, as hexamer frequencies among several species (e.g., mouse, rhesus and human) show minimal variability (data not shown). Third, locating these rare hexamers may be difficult in small target transcripts, and thus other means to limit off-targeting may be necessary. For instance, several reports have demonstrated that certain chemical modifications to the seed nucleotides significantly reduce off-targeting from chemically synthesized siRNAs.36,37,38 The prospect of combining “safe” seed design with chemical modifications serves as a provocative strategy to develop synthetic siRNAs with very high specificity. However, for expressed RNAi, chemical modifications are not applicable, thus “safe” seed design provides the primary means to limit off-targeting for these hairpin-based vectors.

In summary, “safe” seed siRNA design has significant implications for therapeutic development which may result in substantial time- and cost-saving opportunities. Traditional small molecules are initially screened for efficacy and later tested for safety, since predicting potential side effects remains a challenge due to the complex nature of small molecule interactions. By contrast, our ability to predict off-targeting (derived from base-pairing) for oligonucleotide-based drugs provides a unique opportunity to prioritize safety during drug development and subsequently screen for efficacy. Using this reverse paradigm, researchers improve their chances to identify lead candidates with a highly desirable balance of potency and specificity, increasing their likelihood for success in preclinical studies and early stage clinical trials.

Materials and Methods

Plasmids and viral vectors. The plasmids expressing mouse U6-driven artificial miRNAs were cloned as previously described using the DNA oligonucleotides listed in Supplementary Table S3.27 For AAV production, artificial miRNA expression cassettes were cloned into pFBGR-derived plasmids which coexpress CMV-driven green fluorescent protein.29

Recombinant AAV serotype 2/1 vectors (AAV1-GFP, AAV1-miSCR, AAV1-miHDS1, and AAV1-miHDS2) were generated by the University of Iowa Vector Core facility as previously described.39 Viruses were initially purified using an iodixanol gradient (15–60% wt/vol) and subjected to additional purification via ion exchange using MustangQ Acrodisc membranes (Pall, East Hills, NY). AAV1 vectors were resuspended in Formulation Buffer 18 (HyClone, Logan, UT), and titers (viral genomes/ml) were determined by QPCR.

AAV injections and brain tissue isolation. All animal protocols were approved by the University of Iowa Animal Care and Use Committee. Wild-type FVB mice were injected with AAV1 vectors as previously reported.22,28 For all studies, unless indicated otherwise, mice were injected bilaterally into the striatum (coordinates: 0.86 mm rostral to bregma, ±1.8 mm lateral to midline, 3.5 mm ventral to the skull surface) with 4 µl of AAV1 virus (at ~1 × 1012 viral genomes/ml). Mice used in histological analyses were anesthetized with a ketamine/xylazine mix and transcardially perfused with 20 ml of 0.9% cold saline, followed by 20 ml of 4% paraformaldehyde in 0.1 mol/l PO4 buffer. Mice were decapitated, and the brains were removed and post-fixed overnight in 4% paraformaldehyde. Brains were stored in a 30% sucrose solution at 4 °C until cut on a sliding knife microtome at 40 µm thickness and stored at –20 °C in a cryoprotectant solution. Mice used for QPCR analyses were perfused with 20 ml of 0.9% cold saline. Brains were removed and sectioned into 1-mm thick coronal slices using a brain matrix (Roboz, Gaithersburg, MD). Tissue punches were taken from the striatum using a tissue core (1.4 mm in diameter) and triturated in 50 µl of TRIzol (Invitrogen, Carlsbad, CA). RNA was isolated from striatal punches using 1 ml of TRIzol.

Immunohistochemical analyses. Free-floating, coronal brain sections (40 µm thick) were processed for immunohistochemical visualization of microglia (anti-Iba1, 1:1,000; WAKO, Richmond, VA). All staining procedures were carried out as previously described,28 using goat anti-rabbit IgG secondary antibody (1:200) and Vectastain ABC-peroxidase reagent (both from Vector Laboratories, Burlingame, CA). Stained or unstained (the latter for visualization of green fluorescent protein autofluorescence) sections were mounted onto Superfrost Plus slides (Fisher Scientific, Pittsburgh, PA) and coverslipped with Gelmount (Biomeda, Foster City, CA) or Vectashield (Vector Laboratories). Images were captured using an Olympus BX60 light microscope and DP70 digital camera, along with Olympus DP Controller software (Olympus, Melville, NY).

Hexamer distribution analyses. All human RefSeq IDs, official gene symbols, and coding and 3′-UTR sequences (Hg19, GRCH37) were obtained and only sequences with NM_* prefixes were used for analysis. For 3′-UTR sequences, the nonoverlapping frequency of each individual hexamer (4,096 possible) was counted to determine the number of 3′-UTRs containing a given hexamer (Supplementary Table S2). Nonoverlapping sites were considered to account for actual binding site availability. For coding sequence, the total hexamer frequencies were determined, allowing overlapping hexamers, to estimate the probability of selecting siRNA sequences containing the specified hexamer. For genes with variants (i.e., same official gene symbol but different accession number), the maximum count for each hexamer was used.

Cell culture and transfection. For the HDS screen, HEK293 cells were grown in 24-well plates in growth media containing 10% fetal bovine serum and transfected in quadruplicate with 400 ng of plasmid using Lipofectamine 2000 (Invitrogen) by adding the lipid:DNA complexes directly to the growth media. Total RNA was isolated at 24 hours post-transfection using 1 ml of Trizol. For microarray studies, HEK293 cells were grown in 12-well plates in growth media (10% fetal bovine serum) and transfected with 1 µg of plasmid under serum-free conditions. Lipid:DNA complexes were removed 3 hours later and replaced with growth media (5% fetal bovine serum). At 72 hours (microarray) post-transfection, total RNA was isolated using 1 ml of TRIzol.

Quantitative real-time PCR (QPCR). Random-primed first-strand complementary DNA (cDNA) synthesis was performed using 500 ng total RNA (High Capacity cDNA Reverse Transcription Kit; Applied Biosystems, Foster City, CA) per manufacturer's protocol. Assays were performed on a sequence detection system using primers/probe sets specific for human htt and GAPDH or mouse htt and β-actin (Prism 7900HT and TaqMan 2X Universal Master Mix; Applied Biosystems). Relative gene expression was determined using the ΔΔCT method, normalizing to either GAPDH or β-actin mRNA levels.

Microarray analyses. Microarray analysis was done with assistance from the University of Iowa DNA Facility (Iowa City, IA). Fifty nanograms of total RNA template were used to produce amplified cDNA using the Ovation Biotin RNA Amplification System, v2 (NuGEN Technologies, San Carlos, CA) following the manufacturer's protocol. Amplified cDNA product was purified with DNA Clean and Concentrator-25 (Zymo Research, Irvine, CA). 3.75 µg of amplified cDNA were processed using the FL-Ovation cDNA Biotin Module v2 (NuGEN Technologies) to produce biotin labeled antisense cDNA in 50–100 base pair fragments. Following denaturation at 99 °C for 2 minutes, fragmented, labeled cDNA were combined with hybridization control oligomer (b2) and control cRNAs (BioB, BioC, BioD, and CreX) in hybridization buffer and hybridized to the HuGene 1.0ST GeneChip (Affymetrix, Santa Clara, CA) capable of detecting >28,000 genes. Following an 18-hour incubation at 45 °C, the arrays were washed, stained with streptavidinphycoerythrin (Molecular Probes), and then amplified with an anti-streptavidin antibody (Vector Laboratories) using the Fluidics Station 450 (Affymetrix). Arrays were scanned with the Affymetrix Model 3000 scanner and data collected using GeneChip operating software (GCOS) v1.4. Each sample and hybridization underwent a quality control evaluation, including percentage of probe sets reliably detecting between 40 and 60% present call and 3′–5′ ratio of the GAPDH gene <3.

Partek Genomics Suite (Partek GS, St Louis, MO) was used to preprocess, normalize and analyze microarray data. Affymetrix array raw fluorescence intensity measures of gene expression were normalized and quantified using robust multiarray analysis. To identify differentially expressed genes among the nine treatment groups (N = 4 each, Table 1), we employed two-way ANOVA (variables: scan date and treatment) because arrays were processed in groups of four (one replicate per treatment in each group). Pair-wise contrasts between groups of interest were performed when indicated. Principal component and hierarchical clustering analyses were used to visualize differential gene expression.

Cumulative distribution Analyses. 3′-UTR sequences for all RefSeq mRNAs on the HuGene 1.0 ST chip were obtained, and the number of nonoverlapping seed complement binding sites (octamers) per 3′-UTR for each of the indicated inhibitory RNAs was determined. Three possible octamers for each artificial miRNA were considered to account for flexibility in Drosha and Dicer cleavage (Supplementary Table S3). Transcripts were parsed into groups depending on the number of seed complements in their 3′-UTR (no sites, 1 site, 2+ sites), and cumulative distributions of gene expression values (Log2 fold-change, relative to U6) were plotted. Two-sample Kolmogorov–Smirnov tests were performed to evaluate the statistical significance of distributional deviations relative to baseline (no sites).

Motif discovery. The Venn diagram feature on Partek GS was used to create lists of uniquely downregulated genes (1.1-fold, P < 0.05 or 1.2-fold, P < 0.01, relative to U6) for each treatment, taking into account htt silencing (e.g., HDS1, HDS2, HD2.4, and HD8.2 were included in one Venn diagram, and Safe, Terror, 8.2mis and 8.2-124a were included in another). Ensembl Gene IDs were obtained using the Gene ID Conversion Tool at the David Bioinformatics Resources web-server.40 Ensemble Gene IDs were subjected to target set analysis using the Amadeus Motif Discovery Platform (Allegro Software Package) to identify 8mers enriched in the target set 3′-UTRs, relative the provided human 3′-UTR background dataset.41 Amadeus blindly identified an enrichment of seed complement motifs for each RNAi sequence tested, and the lowest P values for the relevant motifs were reported.

SUPPLEMENTARY MATERIAL Figure S1 Evaluation of microarray data for off-targeting. Table S1 siRNA literature survey. Table S2 Human 3'-UTR hexamer distributions. Table S3 Artificial miRNA sequences.

Acknowledgments

We thank the B.L.D. and McCray laboratories for feedback and discussion, Alex Mas Monteys for sharing his in vivo data for miHD-Ex1, and Ines Martins for assistance with animal injections and histological analyses. This research was supported by funds from the NIH (NS-50210, NS-068099, DK-54759), the Hereditary Disease Foundation, and the Roy J. Carver Trust. R.L.B. is supported by the Lori C. Sasser Fellowship.

Supplementary Material

Figure S1

Evaluation of microarray data for off-targeting.

Click here for additional data file. (957.6KB, pdf)
Table S1

siRNA literature survey.

Click here for additional data file. (315KB, doc)
Table S2

Human 3'-UTR hexamer distributions.

Click here for additional data file. (510.5KB, xls)
Table S3

Artificial miRNA sequences.

Click here for additional data file. (45KB, doc)

REFERENCES

  1. Provost P, Dishart D, Doucet J, Frendewey D, Samuelsson B., and, Rådmark O. Ribonuclease activity and RNA binding of recombinant human Dicer. EMBO J. 2002;21:5864–5874. doi: 10.1093/emboj/cdf578. [DOI] [PMC free article] [PubMed] [Google Scholar]
  2. Krol J, Loedige I., and, Filipowicz W. The widespread regulation of microRNA biogenesis, function and decay. Nat Rev Genet. 2010;11:597–610. doi: 10.1038/nrg2843. [DOI] [PubMed] [Google Scholar]
  3. Lee Y, Ahn C, Han J, Choi, H, Kim, J, Yim, J.et al. (2003The nuclear RNase III Drosha initiates microRNA processing Nature 425415–419. [DOI] [PubMed] [Google Scholar]
  4. Gregory RI, Yan KP, Amuthan G, Chendrimada T, Doratotaj B, Cooch N.et al. (2004The Microprocessor complex mediates the genesis of microRNAs Nature 432235–240. [DOI] [PubMed] [Google Scholar]
  5. Schwarz DS, Hutvágner G, Du T, Xu Z, Aronin N., and, Zamore PD. Asymmetry in the assembly of the RNAi enzyme complex. Cell. 2003;115:199–208. doi: 10.1016/s0092-8674(03)00759-1. [DOI] [PubMed] [Google Scholar]
  6. Khvorova A, Reynolds A., and, Jayasena SD. Functional siRNAs and miRNAs exhibit strand bias. Cell. 2003;115:209–216. doi: 10.1016/s0092-8674(03)00801-8. [DOI] [PubMed] [Google Scholar]
  7. Guo H, Ingolia NT, Weissman JS., and, Bartel DP. Mammalian microRNAs predominantly act to decrease target mRNA levels. Nature. 2010;466:835–840. doi: 10.1038/nature09267. [DOI] [PMC free article] [PubMed] [Google Scholar]
  8. Lewis BP, Burge CB., and, Bartel DP. Conserved seed pairing, often flanked by adenosines, indicates that thousands of human genes are microRNA targets. Cell. 2005;120:15–20. doi: 10.1016/j.cell.2004.12.035. [DOI] [PubMed] [Google Scholar]
  9. Davidson BL., and, McCray PB., Jr Current prospects for RNA interference-based therapies. Nat Rev Genet. 2011;12:329–340. doi: 10.1038/nrg2968. [DOI] [PMC free article] [PubMed] [Google Scholar]
  10. Paul CP, Good PD, Winer I., and, Engelke DR. Effective expression of small interfering RNA in human cells. Nat Biotechnol. 2002;20:505–508. doi: 10.1038/nbt0502-505. [DOI] [PubMed] [Google Scholar]
  11. Zeng Y, Wagner EJ., and, Cullen BR. Both natural and designed micro RNAs can inhibit the expression of cognate mRNAs when expressed in human cells. Mol Cell. 2002;9:1327–1333. doi: 10.1016/s1097-2765(02)00541-5. [DOI] [PubMed] [Google Scholar]
  12. Chi JT, Chang HY, Wang NN, Chang DS, Dunphy N., and, Brown PO. Genomewide view of gene silencing by small interfering RNAs. Proc Natl Acad Sci USA. 2003;100:6343–6346. doi: 10.1073/pnas.1037853100. [DOI] [PMC free article] [PubMed] [Google Scholar]
  13. Semizarov D, Frost L, Sarthy A, Kroeger P, Halbert DN., and, Fesik SW. Specificity of short interfering RNA determined through gene expression signatures. Proc Natl Acad Sci USA. 2003;100:6347–6352. doi: 10.1073/pnas.1131959100. [DOI] [PMC free article] [PubMed] [Google Scholar]
  14. Jackson AL, Bartz SR, Schelter J, Kobayashi SV, Burchard J, Mao M.et al. (2003Expression profiling reveals off-target gene regulation by RNAi Nat Biotechnol 21635–637. [DOI] [PubMed] [Google Scholar]
  15. Birmingham A, Anderson EM, Reynolds A, Ilsley-Tyree D, Leake D, Fedorov Y.et al. (20063' UTR seed matches, but not overall identity, are associated with RNAi off-targets Nat Methods 3199–204. [DOI] [PubMed] [Google Scholar]
  16. Fedorov Y, Anderson EM, Birmingham A, Reynolds A, Karpilow J, Robinson K.et al. (2006Off-target effects by siRNA can induce toxic phenotype RNA 121188–1196. [DOI] [PMC free article] [PubMed] [Google Scholar]
  17. Anderson EM, Birmingham A, Baskerville S, Reynolds A, Maksimova E, Leake D.et al. (2008Experimental validation of the importance of seed complement frequency to siRNA specificity RNA 14853–861. [DOI] [PMC free article] [PubMed] [Google Scholar]
  18. Birmingham A, Anderson E, Sullivan K, Reynolds A, Boese Q, Leake D.et al. (2007A protocol for designing siRNAs with high functionality and specificity Nat Protoc 22068–2078. [DOI] [PubMed] [Google Scholar]
  19. Naito Y, Yoshimura J, Morishita S., and, Ui-Tei K. siDirect 2.0: updated software for designing functional siRNA with reduced seed-dependent off-target effect. BMC Bioinformatics. 2009;10:392. doi: 10.1186/1471-2105-10-392. [DOI] [PMC free article] [PubMed] [Google Scholar]
  20. Jackson AL., and, Linsley PS. Recognizing and avoiding siRNA off-target effects for target identification and therapeutic application. Nat Rev Drug Discov. 2010;9:57–67. doi: 10.1038/nrd3010. [DOI] [PubMed] [Google Scholar]
  21. Ross CA., and, Tabrizi SJ. Huntington's disease: from molecular pathogenesis to clinical treatment. Lancet Neurol. 2011;10:83–98. doi: 10.1016/S1474-4422(10)70245-3. [DOI] [PubMed] [Google Scholar]
  22. Harper SQ, Staber PD, He X, Eliason SL, Martins IH, Mao Q.et al. (2005RNA interference improves motor and neuropathological abnormalities in a Huntington's disease mouse model Proc Natl Acad Sci USA 1025820–5825. [DOI] [PMC free article] [PubMed] [Google Scholar]
  23. Rodriguez-Lebron E, Denovan-Wright EM, Nash K, Lewin AS., and, Mandel RJ. Intrastriatal rAAV-mediated delivery of anti-huntingtin shRNAs induces partial reversal of disease progression in R6/1 Huntington's disease transgenic mice. Mol Ther. 2005;12:618–633. doi: 10.1016/j.ymthe.2005.05.006. [DOI] [PMC free article] [PubMed] [Google Scholar]
  24. Franich NR, Fitzsimons HL, Fong DM, Klugmann M, During MJ., and, Young D. AAV vector-mediated RNAi of mutant huntingtin expression is neuroprotective in a novel genetic rat model of Huntington's disease. Mol Ther. 2008;16:947–956. doi: 10.1038/mt.2008.50. [DOI] [PMC free article] [PubMed] [Google Scholar]
  25. Boudreau RL, McBride JL, Martins I, Shen S, Xing Y, Carter BJ.et al. (2009Nonallele-specific silencing of mutant and wild-type huntingtin demonstrates therapeutic efficacy in Huntington's disease mice Mol Ther 171053–1063. [DOI] [PMC free article] [PubMed] [Google Scholar]
  26. Drouet V, Perrin V, Hassig R, Dufour N, Auregan G, Alves S.et al. (2009Sustained effects of nonallele-specific Huntingtin silencing Ann Neurol 65276–285. [DOI] [PubMed] [Google Scholar]
  27. Boudreau RL, Monteys AM., and, Davidson BL. Minimizing variables among hairpin-based RNAi vectors reveals the potency of shRNAs. RNA. 2008;14:1834–1844. doi: 10.1261/rna.1062908. [DOI] [PMC free article] [PubMed] [Google Scholar]
  28. McBride JL, Boudreau RL, Harper SQ, Staber PD, Monteys AM, Martins I.et al. (2008Artificial miRNAs mitigate shRNA-mediated toxicity in the brain: implications for the therapeutic development of RNAi Proc Natl Acad Sci USA 1055868–5873. [DOI] [PMC free article] [PubMed] [Google Scholar]
  29. Boudreau RL, Martins I., and, Davidson BL. Artificial microRNAs as siRNA shuttles: improved safety as compared to shRNAs in vitro and in vivo. Mol Ther. 2009;17:169–175. doi: 10.1038/mt.2008.231. [DOI] [PMC free article] [PubMed] [Google Scholar]
  30. Matveeva O, Nechipurenko Y, Rossi L, Moore B, Saetrom P, Ogurtsov AY.et al. (2007Comparison of approaches for rational siRNA design leading to a new efficient and transparent method Nucleic Acids Res 35e63. [DOI] [PMC free article] [PubMed] [Google Scholar]
  31. Huesken D, Lange J, Mickanin C, Weiler J, Asselbergs F, Warner J.et al. (2005Design of a genome-wide siRNA library using an artificial neural network Nat Biotechnol 23995–1001. [DOI] [PubMed] [Google Scholar]
  32. Grimson A, Farh KK, Johnston WK, Garrett-Engele P, Lim LP., and, Bartel DP. MicroRNA targeting specificity in mammals: determinants beyond seed pairing. Mol Cell. 2007;27:91–105. doi: 10.1016/j.molcel.2007.06.017. [DOI] [PMC free article] [PubMed] [Google Scholar]
  33. Grimm D, Streetz KL, Jopling CL, Storm TA, Pandey K, Davis CR.et al. (2006Fatality in mice due to oversaturation of cellular microRNA/short hairpin RNA pathways Nature 441537–541. [DOI] [PubMed] [Google Scholar]
  34. Huang C, Li M, Chen C., and, Yao Q. Small interfering RNA therapy in cancer: mechanism, potential targets, and clinical applications. Expert Opin Ther Targets. 2008;12:637–645. doi: 10.1517/14728222.12.5.637. [DOI] [PubMed] [Google Scholar]
  35. Burchard J, Jackson AL, Malkov V, Needham RH, Tan Y, Bartz SR.et al. (2009MicroRNA-like off-target transcript regulation by siRNAs is species specific RNA 15308–315. [DOI] [PMC free article] [PubMed] [Google Scholar]
  36. Jackson AL, Burchard J, Leake D, Reynolds A, Schelter J, Guo J.et al. (2006Position-specific chemical modification of siRNAs reduces “off-target” transcript silencing RNA 121197–1205. [DOI] [PMC free article] [PubMed] [Google Scholar]
  37. Bramsen JB, Pakula MM, Hansen TB, Bus C, Langkjær N, Odadzic D.et al. (2010A screen of chemical modifications identifies position-specific modification by UNA to most potently reduce siRNA off-target effects Nucleic Acids Res 385761–5773. [DOI] [PMC free article] [PubMed] [Google Scholar]
  38. Vaish N, Chen F, Seth S, Fosnaugh K, Liu Y, Adami R.et al. (2011Improved specificity of gene silencing by siRNAs containing unlocked nucleobase analogs Nucleic Acids Res 391823–1832. [DOI] [PMC free article] [PubMed] [Google Scholar]
  39. Urabe M, Ding C., and, Kotin RM. Insect cells as a factory to produce adeno-associated virus type 2 vectors. Hum Gene Ther. 2002;13:1935–1943. doi: 10.1089/10430340260355347. [DOI] [PubMed] [Google Scholar]
  40. Huang da W, Sherman BT, Tan Q, Collins JR, Alvord WG, Roayaei J.et al. (2007The DAVID Gene Functional Classification Tool: a novel biological module-centric algorithm to functionally analyze large gene lists Genome Biol 8R183. [DOI] [PMC free article] [PubMed] [Google Scholar]
  41. Halperin Y, Linhart C, Ulitsky I., and, Shamir R. Allegro: analyzing expression and sequence in concert to discover regulatory programs. Nucleic Acids Res. 2009;37:1566–1579. doi: 10.1093/nar/gkn1064. [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Figure S1

Evaluation of microarray data for off-targeting.

Click here for additional data file. (957.6KB, pdf)
Table S1

siRNA literature survey.

Click here for additional data file. (315KB, doc)
Table S2

Human 3'-UTR hexamer distributions.

Click here for additional data file. (510.5KB, xls)
Table S3

Artificial miRNA sequences.

Click here for additional data file. (45KB, doc)

Articles from Molecular Therapy are provided here courtesy of The American Society of Gene & Cell Therapy

RESOURCES