Local RNA Target Structure Influences siRNA Efficacy: A Systematic Global Analysis

doi:10.1016/j.jmb.2005.03.012

J. Mol. Biol. (2005) 348, 871–881

Local RNA Target Structure Influences siRNA Efficacy: A Systematic Global Analysis Marita Overhoff1†, Martina Alken2†, Rosel Kretschmer-Kazemi Far1 Marc Lemaitre3, Bernard Lebleu2, Georg Sczakiel1* and Ian Robbins2 1

Universita¨t zu Lu¨beck, Institut fu¨r Molekulare Medizin Ratzeburger Allee 160, D-23538 Lu¨beck, Germany

2 UMR 5124 CNRS, Laboratoire des De´fenses Antivirales et Antitumorales, Universite´ Montpellier II, 34095 Montpellier Cedex 5, France 3

EUROGENTEC S.A., Lie`ge Science Park, B-4102 Seraing Belgium

The efficiency with which small interfering RNAs (siRNAs) down-regulate specific gene expression in living cells is variable and a number of sequence-governed, biochemical parameters of the siRNA duplex have been proposed for the design of an efficient siRNA. Some of these parameters have been clearly identified to influence the assembly of the RNA-induced silencing complex (RISC), or to favour the sequence preferences of the RISC endonuclease. For other parameters, it is difficult to ascertain whether the influence is a determinant of the siRNA per se, or a determinant of the target RNA, especially its local structural characteristics. In order to gain an insight into the effects of local target structure on the biological activity of siRNA, we have used large sets of siRNAs directed against local targets of the mRNAs of ICAM-1 and survivin. Target structures were classified as accessible or inaccessible using an original, iterative computational approach and by experimental RNase H mapping. The effectiveness of siRNA was characterized by measuring the IC50 values in cell culture and the maximal extent of target suppression. Mean IC50 values were tenfold lower for accessible local target sites, with respect to inaccessible ones. Mean maximal target suppression was improved. These data illustrate that local target structure does, indeed, influence the activity of siRNA. We suggest that local target screening can significantly improve the hit rate in the design of biologically active siRNAs. q 2005 Elsevier Ltd. All rights reserved.

*Corresponding author

Keywords: RNA structure; siRNA, design; siRNA, mechanism

Introduction RNA interference is an evolutionarily conserved process in which short regions of double-stranded RNA are produced endogenously from larger precursors and induce the post-transcriptional inhibition of the expression of specific genes.1,2 The experimental introduction of such small interfering RNA (siRNA) into eukaryotic cells is rapidly becoming one of the most important new tools in functional genomics and holds great potential as a therapeutic strategy.3 The efficiency with which siRNA down-regulates specific gene expression in living cells is highly variable. Indeed, large-scale screens of a number of † M.O. & M.A. contributed equally to this work. Abbreviations used: siRNA, small interfering RNA; RISC, RNA-induced silencing complex; asRNAi, antisense siRNA strand; FCS, fetal calf serum. E-mail address of the corresponding author: [email protected]

mRNAs have highlighted extensive positional effects, and some siRNAs show limited efficiency.4–6 This suggests that the efficacy of a specific siRNA species is influenced by intrinsic properties of the siRNA itself, by intrinsic properties of the target system (such as target RNA structure) or by a combination of both. A number of thermodynamic and structural characteristics of the siRNA moiety have been suggested to contribute to the effectiveness of siRNA.7,8 Recent studies have led to the establishment of a list of eight such parameters that, when incorporated into a rational design algorithm, permit an increase in the probability of selecting an effective siRNA (i.e. one that is capable of downregulating gene expression O50%.6 This leads to the suggestion that siRNA could be effective independently of its target sequence. However, although it is becoming clear that certain of the biochemical characteristics of the siRNA duplex influence the probability of the antisense siRNA strand (asRNAi) being recruited by the RNA-induced silencing complex (RISC),7,8 it is not yet possible to establish

0022-2836/$ - see front matter q 2005 Elsevier Ltd. All rights reserved.

872 for others whether the critical parameter affected is siRNA-dependent (i.e. RISC formation), targetdependent (i.e. sequence- or structure-dependent), or asRNAi/target duplex-dependent (i.e. substrate for endonuclease activity). A number of reports have suggested that the selection of local target sequences is critical,4,5,9 as is the local structural target accessibility.4,5,9–11 The complex secondary structure of mRNAs, however, is not amenable to accurate modelling,12–15 making it difficult to correlate siRNA efficacy to target structure. By narrowing the definition of secondary structure to two broad categories, accessible or nonaccessible (vis a` vis to Watson–Crick binding) and using an established, iterative computational analysis,16,17 relevant structural information can be gleaned from primary sequence. This kind of theoretical approach has been demonstrated to give results that are highly consistent with structural probing of RNA in the presence of cellular extracts,18 suggesting that it represents a valid basis to study the relationship between target accessibility and efficacy of siRNA. In order to better define the role of local target structure in siRNA efficacy, we have estimated IC50 values for siRNAs targeted to regions defined, by this computational analysis, as structurally accessible, or inaccessible, for two biologically distant target systems ICAM-1 and survivin. In addition, for the survivin transcript, we have used combinatorial oligonucleotide–RNase H mapping to confirm computed accessibility. We demonstrate that siRNA targeting putative accessible sites have a higher probability of exhibiting low IC50 and show greater maximal inhibition than those targeting putative inaccessible sites. Our data are in line with the accompanying study by Schubert et al. (accompanying paper) using intentionally designed target structures to study the relationship between local accessibility and efficacy.31 This suggests that target structure does, indeed, influence siRNA activity.

Results The ICAM-1 and survivin targets: structural analysis and selection of local target sites of mRNA The expression of ICAM-1 in ECV304 cells and the expression of survivin in K562 cells represent two independent biological systems for this study. For both systems, experimental protocols, including cell culture conditions and transfection protocols, have been optimized. Target sites along the ICAM-1 and survivin mRNAs were analysed by a systematic computational calculation of local RNA secondary structures and classified according to their structural accessibility as described.11,16 Briefly, the target sequence is dissected into overlapping windows of a given length (e.g. 800 nt and overlaps of 100 nt) and secondary structures of those windows are calculated. Subsequently, we record

Target Structure and siRNA Efficacy

stretches of consecutively unpaired nucleotides (R10 nt) and their abundance in all predictions. We regard as accessible the target segments that show the longest and most highly conserved consecutively unpaired sequence segments. In principle, this computational analysis reveals nonpaired segments rather than specific folding units. This means that this study does not rely on the relevance of predicted RNA structures but on the meaning of the statistical prediction of sequence segments of the target RNA that are not involved in intramolecular folding and, thus are regarded as accessible in this work. This approach was shown earlier to give rise to results that are consistent with a relevant protocol for the experimental probing of target RNA in the presence of cellular extracts.18 Structurally accessible and inaccessible local target sites were defined according to this concept for the full-length spliced mRNAs of ICAM-1 and survivin as summarized in Figure 1. Design of siRNA The analysis of local structural target accessibility shown in Figure 1 was used to define the sequences of siRNA such that the antisense strand of the selected siRNA was complementary to an accessible or an inaccessible stretch of target sequences. Further, target sites were distributed along the entire target RNA (Figure 1 and Table 1). In addition to local structural characteristics of the target RNA the design of siRNA sequences in this work was based on criteria related to their sequences including nucleotide sequence homology with human sequences and, with one exception (see below), the exclusion of specific motifs such as G-quartets. All siRNA constructs contained a 21 bp doublestranded portion and were tested for nucleotide sequence homology with unrelated human transcribed sequences. The controls with scrambled nucleotide sequences (see Table 1) were designed such that they cover a range of GC content (9–15 nt) and consequent thermal stability. Efficacy of ICAM-1 and survivin-directed siRNA in transfected cells Suppression of ICAM-1 gene expression was measured by transient transfection of ECV304 cells with siRNA followed by stimulation of ICAM-1 gene expression by IL-1b.11 In K562 cells, survivin-directed siRNA was transfected by electroporation because we observed that these cells showed an increase of survivin expression with the use of most transfecting agents (data not shown). Since the action of siRNA occurs post-transcriptionally, siRNA-mediated effects were monitored at the target RNA level by quantitative RT-PCR. In order to compare the efficacy of all siRNA, two parameters were derived, the calculated halfmaximal inhibitory concentration (IC50) and the maximal extent of target suppression within the tested range. Values were standardized to the

Target Structure and siRNA Efficacy

873

Figure 1. Accessibility of local target sites of ICAM-1 mRNA (top panel) and survivin mRNA (lower panel). Sites predicted to be accessible by iterative computational analysis are indicated by dark green arrows, those predicted to be inaccessible are indicated by red arrows. Local target sites identified as accessible by RNase H mapping are indicated by light green arrows, those that were found to be inaccessible by this technique are indicated by black arrows. Note that the site 729 of surviving mRNA is subdivided into 729-1 and 729-3 (see also Figure 3). Abbreviations: 5 0 -UTR, 5 0 untranslated region; cds, coding sequence; 3 0 -UTR, 3 0 untranslated region.

IL-1b-induced state in the case of ICAM-1 expression (Table 2) and to the expression of survivin in the presence of an unrelated siRNA against EGFP which had no effect on survivin expression in K562 (cf. non-treated cells; Table 3). In the context of the IC50 values, in the ICAM-1 system there is clearly a more efficient target suppression by siRNAs directed against the accessible target sites, with a mean value of 4 nM (Table 2A). The siRNAs targeted to the inaccessible sites, on the other hand, were tenfold less effective, with a mean IC50 value of 42 nM (Table 2B). This major difference between the two groups can be accounted for by the greater abundance of less effective (or even non-effective) siRNA species within the group directed against inaccessible targets, although this group includes species that exert high efficacy, e.g. si640, si650, or si1437 (Table 2B). In contrast, all of the siRNAs directed against accessible local target sites were effective, as reflected by their low IC50 values (in the 1–10 nM range). Some species (si862, si1628, and si1637) showed equally strong inhibition between 1 nM and 6 nM, though with a high error range. These data are, therefore, excluded from Table 2. A similar relationship between the computational analysis of target site accessibility and the effectiveness of siRNA directed against survivin was observed in K562 cells (Table 3). Although a smaller number of cases was evaluated, and generally higher IC50 values were calculated, the improvement of the efficacy when considering the structural analysis of local targets was reflected here by an 11-fold improvement of the mean value for IC50 for the accessible target group. The major difference between the accessible and inaccessible target groups is the proportion of siRNA species with a low IC 50. All siRNAs

targeting accessible sites of ICAM-1 and survivin have an IC50%10 nM, whereas 50% of ICAM-1targeted siRNAs, and 75% of survivin-targeted siRNAs, in the inaccessible groups had IC50O 10 nM. This dichotomy in the efficacy of siRNA targeting accessible and inaccessible groups is reflected also by the extent of inhibition at the maximal dose tested (100 nM for ICAM-1 and 1000 nM for survivin). However, with only one exception, all of the siRNAs, whether directed against accessible or inaccessible local targets, were capable of inducing at least 50% inhibition (Tables 2 and 3). This would suggest that, in principle, the majority of siRNA may be active at higher concentrations. It is noteworthy that a plot of the IC50 values of ICAM-1-directed siRNA versus the position of local targets sites along the ICAM-1 mRNA indicates a site-related rather than siRNArelated correlation with biological activity for the three most upstream located targets (Figure 2). Interestingly, a number of siRNAs behaved irregularly (Table 4). In most cases we observed a lack of dose–response relationship. For example, siRNA1141 (ICAM-1) and si729-1 (survivin) showed an inhibition of 50% and 25%, respectively, at all doses tested (up to 1 mM for si729-1; data not shown) and si2304 showed very strong but doseindependent inhibition. In some cases this may be due to structural constraints inherent to the siRNA molecule per se. For instance, si729-1 was directed against a sequence within the survivin transcript (729–744) that we identified as an obvious and highly promising locally accessible target site. This siRNA contains a consecutive stretch of six GC base-pairs (with five consecutive G nucleotides in the sense strand), which may support the formation of higher-order structures. A stimulation of ICAM-1 and survivin expression was observed occasionally

874

Target Structure and siRNA Efficacy

Table 1. Nulceotide sequences and target positions of ICAM-1-directed siRNA and control siRNA used in this study Nomenclature

Target positions

Nucleotide sequence of the sense strand 5 0 /3 0 5

ICAM-1 si-424 si-429 si-640 si-650 si-839 si-840 si-841 si-842 si-843 si-859 si-860 si-861 si-862 si-1141 si-1153 si-1437 si-1452 si-1546 si-1566 si-1591 si-1595 si-1628 si-1637 si-2289 si-2304 survivin si260-1 si323-1 si729-1 si729-3 si745-2 si858-2 si1098-3 si1133-2 si1530-2 Controls si-sc1 si-sc2 si-sc3 si-sc4 si-sc5 EGFP siRNA

0

30

UCUUGGCAGCCAGUGGGCAAG GCAGCCAGUGGGCAAGAACCU CAAGGGCUGGAGCUGUUUGAG AGCUGUUUGAGAACACCUCGG CCACAGUCACCUAUGGCAACG CACAGUCACCUAUGGCAACGA ACAGUCACCUAUGGCAACGAC CAGUCACCUAUGGCAACGACU AGUCACCUAUGGCAACGACUC GACUCCUUCUCGGCCAAGGCC ACUCCUUCUCGGCCAAGGCCU CUCCUUCUCGGCCAAGGCCUC UCCUUCUCGGCCAAGGCCUCA CCAGAGGACAACGGGCGCAGC GGGCGCAGCUUCUCCUGCUCU GAGCACUCAAGGGGAGGUCAC GGUCACCCGCGAGGUGACCGU GGCCUCAGCACGUACCUCUAU UAACCGCCAGCGGAAGAUCAA UACAGACUACAACAGGCCCAA GACUACAACAGGCCCAAAAAG AACCGAACACACAAGCCACGC CACAAGCCACGCCUCCCUGAA GUACACUGCAGGAGAGUGCCU GUGCCUGGCAAAAAGAUCAAA

424 429 640 650 839 840 841 842 843 859 860 861 862 1141 1153 1437 1452 1546 1566 1591 1595 1628 1637 2289 2304

444 459 660 670 869 860 861 862 863 889 880 881 882 1161 1173 1457 1472 1566 1586 1611 1615 1648 1657 2319 2324

CCAUAGAGGAACAUAAAAAGC GUUUGAAGAAUUAACCCUUGG UUUUUGGGGGCUCAUUUUUGC CUGUUUUGAUUCCCGGGCUUA GUGAGAAGUGAGGGAGGAAGA UCAUGUUGUUGAGGCUGUCAC CACAGAAUAGCACAAACUACA GCCAUUCUAAGUCAUUGGGGA GAGCUGCAGGGUGGAUUGUUA

267 325 716 736 761 858 1103 1142 1533

287 345 736 756 781 878 1123 1162 1553

CGAACUCACUGGUCUGACCAU CGAACUCACUGGUCUGACCGC CGGACGCACUGGUCUGACCGG AGAACACACAGGUCUGAACAU CGAACUCACUGGUCUGAC CUACAACAGCCACAACGUCTT

– – – – – –

– – – – – –

after cationic lipid delivery of siRNA (unpublished data) in a way that seems to include nucleotide sequence-specificity. Therefore, one might speculate on dual influences: a suppressive effect by siRNA mechanisms and a stimulatory effect due to unknown effects that might lead to the loss of concentration dependency, perhaps including effects by the short, double-stranded siRNA. If stimulation of the target dominates over siRNAmediated inhibition, one might even observe increased target expression at increasing doses of siRNA. Relationship between accessible sites as defined by RNase H mapping or by computational analysis and efficacy in living cells Occasionally we found biologically effective siRNA designed against local target sites that have not been predicted to be accessible by

computational analysis, e.g. si260-1, si745-2, or si1530-2 (Table 3B). In order to shed light on the possibility that this is due to misleading results of the computational analysis of target RNA or to a complex structure–function relationship, we performed experimental structural probing of these local target sites. Combinatorial in vitro screening has proved useful for the design of antisense oligonucleotides20 but, to our knowledge, this approach has not been applied to the design of siRNA. The two sites clearly identified as inaccessible by computational analysis for the survivin transcript (323 and 858) were clearly inaccessible in the mapping analysis (Figures 1 and 3). Three sites predicted to be accessible (745, 1098, and 1133) were mapped as accessible using the combinatorial libraries. Two sites shown to have clear accessibility in the RNase H mapping (260 and 1530), however, did not show up in the computational analysis. The sites located between nucleotides 715 and 767 of the

875

Target Structure and siRNA Efficacy

Table 2. Suppression of ICAM-1 gene expression by siRNA monitored by the siRNA concentration of half maximal inhibition (IC50) and the maximal extent of target suppression, i.e. the remaining gene expression within the tested range monitored by target gene expression at 100 nM siRNA (I100) for local target sites that are considered to be accessible (A) or inaccessible (B)

A. Accessible si839 si840 si841 si842 si843 si859 si860 si861 si1595 si1546 Mean B. Inaccessible si424 si429 si640 si650 si1153 si1437 si1452 si2289 Mean

IC50 (nM)

I100 nM (%)

5G3 4G2 4G30 3G2 3G2 4G5 2G23 10G15 2G8 4G2 4

22G1 28G1 35G4 20G3 8G8 26G13 33G15 31G24 18G16 11G1 23

O 100 O 80 5G2 4G2 7G18 6G5 40G5 O100 O42

50G30 40G10 25G3 10G2 20G3 30G12 18G14 z100 37

survivin transcript showed generally good agreement between the two methods. However, the computational analysis predicts the accessible region to extend about eight nucleotides further on the 5 0 side than the accessibility mapped with the libraries. This means that there is good agreement for the targets of si729-3 and si745-2, but that si729-1 is predicted to be accessible by computational analysis but inaccessible by mapping analysis. In total, therefore, six sites showed good agreement and three sites did not, suggesting a general coherence between the two methods. Interestingly, of the two sites found to be accessible by RNase H mapping but not by computational analysis, one has high efficiency (1530) and the other moderate efficiency (260). This is probably a reflection of the high stringency of the computational method.

Discussion It is commonly agreed that not all siRNA species designed against a given target are equally effective, and a number of models have been postulated to explain this observation. It has often been assumed that the biological effectiveness of siRNA is independent of characteristics of the target mRNA and is influenced solely by characteristics intrinsic to individual siRNAs. Thus, the first guidelines established, in order to design efficient siRNAs, put greater emphasis on siRNA sequence per se than on target sequence. Recently, experimental evidence has been provided to support the notion that the asymmetric thermodynamic stability of the siRNA

Table 3. Suppression of survivin gene expression by siRNA monitored by the IC50 values and the maximal extent of target suppression, i.e. the remaining gene expression within the tested range, i.e. at 1 mM (1000 nM) of siRNA (I1000) for local target sites that are considered to be accessible (A) or inaccessible (B)

A. Accessible si729-3 si1098-3 si1133-2 Mean B. Inaccessible si260-1 si323-1 si745-2 si1530-2 Mean

IC50 (nM)

I1000 nM (%)

8G6 8G6 10G10 w9

19G12 4G1 16G10 w13

20G10 15G7 350G200 %10 %99

11G5 17G16 31G9 7G4 w15.5

In order to derive IC50 values from the concentration dependence of siRNA-mediated target suppression we assume that at a concentration of 1 nM there is no biological effect.

double-strand is related to efficacy in living cells.7,8 This demonstrates clearly that the biophysical parameters of the siRNA dimer are indeed important contributors to efficacy. This, however, does not exclude the possibility that local target accessibility contributes to activity. Several studies have demonstrated that target accessibility, resulting from sequence-dependent structural constraints, greatly influences the activity of antisense oligonucleotides.21,22 Although the mechanisms of siRNA are very different from those of antisense oligonucleotides (cytoplasmic versus nuclear compartment; RNA sequence-guided nucleoprotein complex binding versus solution binding), it is clear that Watson–Crick base-pairing is central to both effects, suggesting that local target accessibility could be a critical factor in the siRNA effect. Indeed, in a parallel screen, the efficacy of siRNA and RNase H-competent oligonucleotides shows some degree of homology.5,11 Likewise, in a number of other specific cases, local target characteristics seem to be related to the effectiveness of siRNA-driven target suppression.4,9,11 However, it should be noted that the efficacy of siRNA and antisense oligonucleotides against the same local target may vary substantially.11,23 Finally, it should be remembered that neither the experimental nor the theoretical approach to monitor structural accessibility have a 100% hit rate, though the consistency of both independent approaches is remarkable. Here, we describe the relationship between the biological effectiveness of siRNA and local structural characteristics of target (Figure 1; Tables 2 and 3). Secondly, we compare a computational and an experimental strategy to look at target structures and compare how the perceived structures relate to efficacy (Figure 1). It is important to point out that: (i) all siRNA in the inaccessible group did reduce target expression, albeit with lower efficacy for many of them; and (ii) the difference between the

876

Target Structure and siRNA Efficacy

Figure 2. Relationship between the position of local targets along the mRNAs of ICAM-1 (upper panel) and survivin (lower panel), and efficacy of individual siRNA monitored by IC50 values. Sites predicted to be accessible by iterative computational analysis are indicated by dark green arrows, those predicted to be inaccessible are indicated by red arrows. Local target sites identified as accessible by RNase H mapping are indicated by light green arrows, those that were found to be inaccessible by this technique are indicated by black arrows.

two groups is not merely a shift in the normal distribution of the IC50, but rather reflects the appearance of a bimodal distribution within the inaccessible group. In order to determine to what extent local target structural constraints do indeed influence siRNA function, we have screened the complete mRNAs of survivin and ICAM-1, using an original, iterative, computational approach,17 and we have designed a number of siRNAs targeting sites assigned to one of two groups: accessible or inaccessible. The mean biological effectiveness of both survivin-directed and ICAM-1-directed siRNA was about tenfold higher for the group of siRNA targeting sites defined as accessible by this computational approach. This suggests strongly that characteristics of the local targets, including structural constraints, are important factors in the mechanism of action of siRNA in mammalian cells. In keeping with our observations, the kinetics of the interaction between siRNA and its local target sequence has been demonstrated recently to be clearly related to efficacy in mammalian cells.19 That study describes how the kinetic characteristics of siRNA are tightly related to biological activity as well as to local target structure, i.e. there is a consistency between three parameters, the kinetics/ Kd, the accessibility of local target structure, and the biological effectiveness in mammalian cells. The interactions between siRNA and its target RNA might be influenced substantially by the structure of the target RNA and by the local thermodynamic stability of the siRNA helix. However, there is an inconsistency of the structural analysis at positions 260 and 1530 of the survivin mRNA. In both cases, the more stringent theoretical analysis would not have identified both accessible, i.e. favourable local targets, whereas probing by RNase H did. Further, with respect to the generally great inhibitory potency of siRNA (e.g. versus antisense oligonucleotides), it is not surprising that there are some active species in the category of inaccessible local targets. So, this study does not intend to suggest that all siRNAs against inaccessible targets do not work. It is conceivable, therefore, that other characteristics of the siRNA and/or target become more important for inaccessible sites. For example, one could speculate that mRNA-binding factors that prefer specific sites, structures or sequences influence the extent of siRNA-mediated target suppression. One might speculate that, under the

Table 4. Biochemical characteristics of siRNAs that exert irregular effects in cell culture Name si1141 si1566 si1591 si2304 si-sc4 si729-1 si858-2 a

Local target accessibilitya K C C K C K

Tm (8C)

DG (kcal/mol)

Stability of 3 0 -end (kcal/mol)

Stability of 5 0 -end (kcal/mol)

No. GC basepairs

78.8 73.7 69.6 65.3 65.9 65.8 67.5

K47.1 K48.0 K45.5 K42.2 K43.3 K41.8 K44.9

K9.7 K7.0 K9.9 K7.0 K6.8 K8.2 K9.6

K8.2 K12.3 K7.9 K11.7 K8.1 K4.4 K8.1

15 11 10 9 9 9 10

Symbols: C, accessible; K, inaccessible.

Target Structure and siRNA Efficacy

877

Figure 3. RNase H mapping of the survivin mRNA using four semi-random combinatorial oligonucleotide libraries. Each library was comprised by 11-mer gapmers: 5 0 -mmmXNNNmmmm-3 0 (where m is a random 2 0 -O-methyl residue, N is a random DNA residue and X is a fixed DNA residue; A (lane 1), C (lane 2), G (lane 3) or T (lane 4) for each of the four libraries respectively). Lane 0 is a control lacking RNase H (but with an equimolar mix of the four libraries). RNase H cleavage sites were localized by primer extension. In parallel, pSurvivin was sequenced with the same primers (righthand four lanes). The vertical bars on the right of each gel picture represent the target sites of this mapping approach according to the colour code used in Figure 2. The bars on the left of each gel picture represent sites predicted to be accessible (dark green) or inaccessible (red) by the computational analysis.

experimental conditions used here, structural accessibility is a critical parameter, though inaccessibility may be compensated by other characteristic properties. Whether a helicase activity is part of the mode of action of siRNA/RISC is a matter of controversy. If this activity resolves folded target RNA, then it is difficult to accept a dependence of the biological efficacy of siRNA on local target structure. Conversely, a good deal of experimental evidence argues against this view. Firstly, the group of Berkhout showed that mutations outside the target segment of HIV-1 nef-directed siRNA led to a substantial decrease of the effectiveness of siRNA. A closer look at local RNA structure indicated that an accessible conformation of the local target segment was converted into an inaccessible one by changes of the RNA sequence outside the target stretch.24 This finding indicates the importance of RNA structure for efficacy and further suggests that characteristics of the siRNA (e.g. asymmetric thermodynamic stability) is not the only explanation of biological activity. Secondly, Schubert et al. (see the accompanying article) who studied a single siRNA directed against a matching target segment that was embedded in a different sequence and structure context, provide strong independent evidence for the view the local target structure matters.31 Thirdly, a recent experimental study showed that there is a striking relationship between the extent of local folding of target RNA (ICAM-1 mRNA) and the efficacy of siRNA in mammalian cells,11 which again is clearly compatible with the view that the structural characteristics of the target do matter with respect to biological activity. In sum, these lines of evidence suggest that local RNA structure is related to the biological activity of siRNA, which does not

indicate an RISC-associated helicase activity that melts RNA structure completely before siRNA is recognised. The location of the local target site with respect to the functional domains of the mRNA (5 0 -untranslated region (UTR), coding sequence, 3 0 -UTR) does not seem to play a major role, since favourable target sites and corresponding effective siRNA species are distributed widely (Figure 1). In addition, thermodynamic parameters of the siRNA duplex were analysed as outlined recently.6 Our data do not totally support the general hypothesis of the importance of an asymmetric design of siRNA. For most of the highly potent ICAM-1-directed siRNAs (e.g. si841, si842, si843, si859, and si860) we found a predicted unfavourable stability of the 3 0 -end of the sense strand (5 0 end of the antisense strand). Conversely, the notion that a low level of local stability of the siRNA duplex between positions 9 and 14 favours efficacy is consistent with our data. Indeed, all siRNA species with a stability equal to, or lower than, K9.6 kcal/mol for this internal sequence stretch show a high biological effectiveness (Table 5). However, it should be noted that we do not find biologically ineffective siRNA species that explicitly fullfil the guidelines proposed by Khvorova et al.8 This might be due to the fact that all of the siRNA studied here show clearly higher values for the thermodynamic stability of the 5 0 -end of the antisense strand of the siRNA than those used by Khvorova et al.8 The data summarized in Table 5 for the survivin system further indicate a consistent relationship between the computational target analysis and the RNase H mapping. The experimental mapping predicts a larger number of accessible sites, whereas the theoretical approach seems to be more stringent.

878

Target Structure and siRNA Efficacy

Table 5. Relationship between the biological activity of the siRNA tested in this study and various biochemical characteristics intrinsic to the sequences of the siRNAs used in this study

Name

Predicted accessibilitya

IC50 (nM)

K K K K K K K K C C C C C C C C C C

100 80 5 4 7 6 40 O100 5 4 4 3 3 4 2 10 4 2

ICAM-1 si424 si429 si640 si650 si1153 si1437 si1452 si2289 si839 si840 si841 si842 si843 si859 si860 si861 si1546 si1595

Name

Stability of 3 0 end (kcal/ mol)

Stability of 5 0 end (kcal/ mol)

No. GC basepairs

76.4 76.8 71.9 69.7 78.7 74.7 79.7 73.9 71.4 69.9 70.3 70.0 70.2 77.3 77.4 77.1 71.2 66.8

K50.4 K50.7 K46.9 K45.3 K52.0 K50.1 K53.2 K48.9 K46.5 K45.9 K46.1 K45.6 K46.0 K51.2 K50.4 K50.6 K46.3 K42.8

K9.0 K8.7 K7.9 K10.0 K10.2 K9.6 K10.3 K11.2 K7.9 K8.4 K9.7 K9.2 K9.6 K12.3 K12.3 K11.6 K7.0 K5.2

K7.8 K11.7 K8.5 K9.8 K12.6 K10.6 K10.5 K8.4 K10.1 K8.7 K8.9 K9.1 K9.1 K9.6 K10.4 K9.7 K8.2 K8.4

13 13 12 11 14 13 15 12 12 11 11 11 11 14 13 14 12 10

Stability of 3 0 end (kcal/ mol)

Stability of 5 0 end (kcal/ mol)

No. GC basepairs

K7.9 K8.5 K8.2 K8.4 K7.9 K12.6 K7.1

K8.0 K6.6 K7.6 K9.4 K8.7 K10.4 K10.3

8 8 10 11 8 10 11

Predicted (left) versus RNase H mapped accessibility (right)a

IC50 (nM)

Tm (8C)

DG (kcal/ mol)

K/C K/K C/C K/C C/C C/C K/C

20 15 8 350 18 10 %10

60.9 60.5 67.6 70.1 61.8 68.5 71.2

K38.8 K38.5 K43.1 K47.0 K40.0 K44.6 K46.9

Survivin si260-1 si323-1 si729-3 si745-2 si1098-3 si1133-2 si1530-2 a

Tm (8C)

DG (kcal/ mol)

Symbols: C, accessible; K, inaccessible.

Conversely, all sites predicted to be accessible by computational analysis have been identified by RNase H mapping. This means that, in principle, computational analysis is sufficient to define the structural constraints of the target RNA that are important for the design of effective siRNA species. In conclusion, in the rational design of siRNA, the hit rate can be increased significantly by including a systematic structural analysis of target RNA.

Materials and Methods

Germany). The sequences of all oligonucleotides are shown in Table 1.

Formation of double-stranded siRNA Complementary RNA strands at a concentration of 10 mM were annealed in 15 mM Hepes (pH 7.4), 50 mM potassium acetate, 1 mM magnesium acetate by heating at 90 8C for two minutes followed by incubation at 37 8C for one hour. The formation of double-stranded RNA was confirmed by semi-denaturing, 12% polyacrylamide gel electrophoresis and staining with Stains-All (Sigma-Aldrich, Deisenhofen, Germany).

Oligonucleotides All oligoribonucleotides used here were synthesized with two 3 0 -deoxythymidine nucleotides and quality assured by PAGE purification and matrix-assisted laser desorption/ionization time-of-flight (MALDI-TOF) mass spectrometry (Eurogentec, Seraing, Belgium). Prior to use, oligonucleotides were quantified by UV absorption spectroscopy and their integrity was controlled on denaturing, 12% (w/v) polyacrylamide gels followed by staining with Stains-All (Sigma-Aldrich, Deisenhofen,

RNA secondary structure prediction To predict secondary structures of target sequences, we used the software mfold version 2.3, which is included in the software Wisconsin Package Version 10.0, Genetics computer Group (GCG).25,26 Characteristic parameters of oligonucleotide sequences were determined using the software Oligo version 3.4, which is based on the work of Rychlik and Rhoads.27

879

Target Structure and siRNA Efficacy

RNase H mapping of the survivin transcript The survivin cDNA was produced from a HeLa cell total RNA preparation by RT-PCR, using the primers 5 0 TTTGGATCCGCCAGATTTGAATCGCGG-3 0 and 5 0 TTTAAGCTTTTTATTTCTCAGGAAC-3 0 , and cloned into the BamHI/HindIII sites of plasmid pGEM-3Zf (Promega, Mannheim, Germany) to give pSurvivin. The cloned sequence was verified by sequencing. RNA was transcribed from pSurvivin using the RiboMAX in vitro transcription kit (Promega, Mannheim, Germany) and purified using the RNeasy mini kit (Qiagen, Hilden, Germany). RNase H mapping was performed essentially as described, using four semi-random nucleotide libraries.28 Each library was comprised by 11-mer gapmers: 5 0 mmmXNNNmmmm-3 0 (where m is a random 2 0 -Omethyl residue, N is a random DNA residue and X is a fixed DNA residue; A,C,G or T for each of the four libraries, respectively). The libraries were synthesized specifically by Eurogentec (Seraing, Belgium). In initial experiments, aimed at optimizing experimental conditions, the RNA transcript was 3 0 -labelled by enzymatic ligation of [32P]pCp (specific activity O3000 Ci/mmol; ICN, Eschwege, Germany) using phage T4 RNA ligase (Amersham, Pharmacia, Biotech). The radio-labelled transcript (0.3 pmol) was incubated with various concentrations of each library and Escherichia coli RNase H (Amersham-Pharmacia Biotech, Freiburg, Germany) in mapping buffer (40 mM Tris–HCl (pH 8), 4 mM MgCl2, 1 mM DTT) in a final volume of 5 ml for 20 minutes at 37 8C. The reaction was stopped by addition of 5 ml of 2! sequencing stop buffer, loaded onto a denaturing, 6% (w/v) polyacrylamide sequencing gel, and analyzed using a PhosphorImager (Molecular Dynamics– Amersham, Freiburg, Germany). Single nucleotide-resolution mapping was performed as described above, using unlabelled survivin transcript and optimized concentrations of oligonucleotide libraries (16 mM) and RNase H (0.1 unit) for the reactions with the C and T libraries; 0.2 unit for the reactions with the A and G libraries). RNase H cleavage sites were localized by primer extension using the AMV Reverse Transcriptase Primer Extension Kit (Promega, Mannheim, Germany). 0.5 pmol of each 20-mer oligonucleotide primer was 5 0 end labelled with 0.25 mCi of [g-32P]ATP (3000 Ci/mmol; ICN, Eschwege, Germany) using one unit of phage T4 polynucleotide kinase (Gibco BRL, Eggenstein, Germany). The labelled primer was hybridized to the RNase H-cleaved transcript at 58 8C for 20 minutes and extended according to the manufacturer’s instructions. In parallel, 2 pmol of the radiolabelled primer was hybridized to 0.5 pmol of denatured (NaOH) pSurvivin and used for manual sequencing with the T7 Sequenase v2.0 kit (USB, Cleveland, USA). For each primer, sequence and mapping reactions were run in parallel on a denaturing, 6% (w/v) polyacrylamide sequencing gel. Data were analysed using a PhosphorImager (Molecular Dynamics–Amersham, Freiburg, Germany). Primers used for primer extension in single nucleotide-resolution mapping experiments were: 5 0 -GAAGGGCCAGTTCTTGAATG-3 0 ; 232–251, 5 0 -CCC AGCCTTCCAGCTCCTTG-3 0 ; 344–363, 5 0 -TCCAGTTT CAAAAATTCACC-3 0 ; 470–489, 5 0 -GGCCAGAGGCCT CAATCCAT-3 0 ; 608–627, 5 0 -AGGAGCACAGTTGAAA CATC-3 0 ; 5 0 -TGAGCCCCCAAAAAAGAGAG-3 0 ; 830– 849, 5 0 -CACATTCACTGTGGAAGGCT-3 0 ; 948–967, 5 0 AAACTGTCCTCTGAGGAGGC-3 0 ; 1045–1064, 5 0 -AAA CAGTAGAGGAGCCAGGG-3 0 ; 1179–1188, 5 0 -ATTCT

GTCTCCTCATCCACC-3 0 ; 1305–1324, 5 0 -AAAGGATT TAGGCCACTGCC-3 0 ; 1422–1441, 5 0 -GGCGGACTGCGT CTCTCCCC-3 0 ; 1538–1557, 5 0 -GCTGTAACAATCCACCC AGC-3 0 ; 1590–1609, 5 0 -GCTTTTTATTTCTCAGGAAC-3 0 . Cell lines and cell culture According to the DSMZ-German Collection of Microorganisms and Cell Cultures (Braunschweig, Germany) the cell-line ECV30429 is a derivative of the human urinary bladder carcinoma cell line T-24, which expresses ICAM-1 in an inducible manner.30 ECV304 cells were maintained in medium 199 (Sigma-Aldrich, Deisenhofen, Germany) buffered with 25 mM Hepes (pH 7.4) and supplemented with 0.68 mM L-glutamine and 10% (v/v) fetal calf serum (FCS; Invitrogen, Karlsruhe, Germany). Cells were split routinely two or three times a week after trypsinization. For stimulation of ICAM-1 (CD54) 200 units/ml of interleukin 1b (IL-1b; PromoCell, Heidelberg, Germany) was added to the medium and cells were incubated overnight (16–18 hours). The survivin-expressing lymphoblast-derived cell line K562 was maintained in suspension culture in RPMI-glutamax medium (Invitrogen, Karlsruhe, Germany) supplemented with 10% FCS (Invitrogen, Karlsruhe, Germany). Cells were passaged every third day (maximum of ten passages). Transfection of cells with siRNA ECV304 cells were seeded in 12-well culture plates at a density of 1.5!105 cells/well, 15 hours prior to oligonucleotide treatment. Then, the cells were washed once with prewarmed (37 8C) phosphate-buffered saline (PBS). To deliver siRNA to cells we used Lipofectamine 2000 (Invitrogen, Karlsruhe, Germany). The transfection of siRNA was performed with 0.4 ml of Opti-MEM I medium (Invitrogen, Karlsruhe, Germany) containing various concentrations of siRNA and 10 mg/ml of Lipofectamine 2000 per well. The cells were incubated for four hours at 37 8C under 5% (v/v) CO2. Subsequently, the transfection medium was replaced by medium 199 containing 10% FCS. After incubation for four hours at 37 8C under 5% CO2, the medium was again substituted with medium 199 containing 10% FCS and 200 units/ml of IL-1b to stimulate ICAM-1 expression overnight. K562 cells passaged 24 hours prior to transfection were recovered by centrifugation, washed in PBS (37 8C) and suspended in nucleofector (Amaxa: kit V, Ko¨ln, Germany) solution (37 8C) at a density of 1!107 cells/ml. Then 2 ml of a 50-fold concentrated siRNA solution was added to 100 ml of cell suspension and transfection carried out according to the Amaxa electroporation protocol specific for these cells. Immediately after transfection, 0.5 ml of pre-warmed complete medium was added to the electroporation cuvette. The cells in their medium were then transferred to a well of a six-well culture dish containing 1.5 ml of complete medium that had been pre-equilibrated to 37 8C under 5% CO2 for several hours. Cells were sacrificed 16 hours post transfection. Quantification of ICAM-1-specific target RNA: quantitative RT-PCR Total RNA was extracted from transfected cells using an RNeasy mini kit, including treatment with RNase-free DNase I (Qiagen, Hilden, Germany). The yield and purity of RNA were determined by spectrophotometry. Synthesis of cDNA was carried out using random hexamer

880 primers and TaqMan reverse transcription reagents according to the manufacturer’s specifications (Applied Biosystems, Darmstadt, Germany). Quantitative PCR was performed using the GeneAmp 5700 sequence detection system (Applied Biosystems, Darmstadt, Germany) and SYBR green PCR core reagents (Eurogentec, Seraing, Belgium). ICAM-1 cDNA was amplified with the forward primer 5 0 -GCCACTTCTTCTGTAAGTCTGTGGG-3 0 and reverse primer 5 0 -CTACCGGCCCTGGGACG-3 0 , resulting in a fragment of 300 bp. Samples were standardized using primers specific to cDNA encoding human GAPDH (forward primer, 5 0 -AACAGCGACACCCACTCCTC-3 0 and reverse primer, 5 0 -GGAGGGGAGATTCAGTGT GGT-3 0 ) resulting in a product of 258 bp. Standard curves were obtained after amplification of 2.5!10 to 2.5!106 copies of purified plasmid pP5, a derivative of pEGFP-C1 carrying the amplicon generated with the ICAM-1 primer set and plasmid pCR-GAPDH, a derivative of pCR 2.1 harbouring the GAPDH amplicon sequences. Quantification of survivin-specific target RNA: quantitative RT-PCR Total RNA was extracted from transfected cells using the RNeasy mini kit, including treatment with RNase-free DNase I (Qiagen, Hilden, Germany). The yield and purity of RNA were determined by spectrophotometry. cDNA was synthesized from the survivin-specific primer 5 0 GGAACAGCCGAGATGACCTCC-3 0 and the S26-specific primer 5 0 -TCTCCAGAGAATAGCCTGTCT-3 0 using Superscript III reverse transcriptase (Invitrogen, Karlsruhe, Germany) according to the manufacturer’s protocol. The cDNA products were purified using the QIAquick PCR purification kit (Qiagen, Hilden, Germany) and quantified by real-time PCR using a lightcycler (Roche, Mannheim, Germany) and SYBR green PCR core reagents (Eurogentec, Seraing, Belgium). The survivin cDNA was amplified with the forward primer 5 0 -GTCAGCCCAACCTTCACATC-3 0 and reverse primer 5 0 -GAAGCTGTAACAATCCACCCT-3 0 , resulting in a fragment of 176 bp. Samples were standardized using primers specific to cDNA encoding human S26 (forward primer, 5 0 -TCTCCGGTCCGTGCCT-3 0 and reverse primer, 5 0 -GCTTCACATACAGCTTGGG-3 0 ) resulting in a product of 244 bp. Standard curves were obtained after amplification of 10K6–10K10 dilutions of PCR products that had been verified by automated sequencing.

Acknowledgements We thank the Medical Faculty of the University of Lu¨beck for financial support within the program FSO to R.K.K.F. and to G.S., and the Association pour la Recherche sur le Cancer to I.R. and to B.L.

References 1. Hannon, G. J. (2002). RNA interference. Nature, 418, 244–251. 2. Mello, C. C. & Conte, D., Jr (2004). Revealing the world of RNA interference. Nature, 431, 338–342. 3. Dorsett, Y. & Tuschl, T. (2004). siRNAs: applications in functional genomics and potential as therapeutics. Nature Rev. Drug. Discov. 3, 318–329.

Target Structure and siRNA Efficacy

4. Holen, T., Amarzguioui, M., Wiiger, M. T., Babaie, E. & Prydz, H. (2002). Positional effects of short interfering RNAs targeting the human coagulation trigger tissue factor. Nucl. Acids Res. 30, 1757–1766. 5. Vickers, T. A., Koo, S., Bennett, C. T., Crooke, S. T., Dean, N. M. & Baker, B. F. (2003). Efficient reduction of target RNAs by small interfering RNA and RNase H-dependent antisense agents. A comparative analysis. J. Biol. Chem. 278, 7108–7118. 6. Reynolds, A., Leake, D., Boese, Q., Scaringe, S., Marshall, W. S. & Khvorova, A. (2004). Rational siRNA design for RNA interference. Nature Biotechnol. 22, 326–330. 7. Schwarz, D. S., Hutva´gner, G., Du, T., Xu, Z., Aronin, N. & Zamore, P. (2003). Asymmetry in the assembly of the RNAi enzyme complex. Cell, 115, 199–208. 8. Khvorova, A., Reynolds, A. & Jayasena, S. D. (2003). Functional siRNAs and miRNAs exhibit strand bias. Cell, 115, 209–216. 9. Bohula, E. A., Salisbury, A. J., Sohail, M., Playford, M. P., Riedemann, J., Southern, E. M. & Macaulay, V. M. (2003). The efficacy of small interfering RNAs targeted to the type 1 IGF receptor is influenced by secondary structure in the IGF1R transcript. J. Biol. Chem. 278, 15991–15997. 10. Xu, Y., Zhang, H.-Y., Thormeyer, D., Larsson, O., Du, Q., Elme´n, J. et al. (2003). Effective small interfering RNAs and phosphorothioate antisense DNAs have different preferences for target sites in the luciferase mRNAs. Biochem. Biophys. Res. Commun. 306, 712–717. 11. Kretschmer-Kazemi Far, R. & Sczakiel, G. (2003). The activity of siRNA in mammalian cells is related to structural target accessibility: a comparison with antisense oligonucleotides. Nucl. Acids Res. 31, 4417–4424. 12. Lima, W. F., Monia, B. P., Ecker, D. J. & Freier, S. M. (1992). Implication of RNA structure on antisense oligonucleotide hybridization kinetics. Biochemistry, 31, 12055–12061. 13. Michel, F. & Westhoff, E. (1990). Modelling of the three-dimensional architecture of group I catalytic introns based on comparative sequence analysis. J. Mol. Biol. 216, 585–610. 14. Stein, C. A. (2001). The experimental use of antisense oligonucleotides: a guide for the perplexed. J. Clin. Invest. 108, 641–644. 15. Sohail, M., Akhtar, S. & Southern, E. M. (1999). The folding of large RNAs studied by hybridization to arrays of complementary oligonucleotides. RNA, 5, 646–655. 16. Patzel, V., Steidl, U., Kronenwett, R., Haas, R. & Sczakiel, G. (1999). A theoretical approach to select effective antisense oligonucleotides at high statistical probability. Nucl. Acids Res. 27, 4328–4334. 17. Kretschmer-Kazemi Far, R., Nedbal, W. & Sczakiel, G. (2001). Concepts to automate the theoretical design of effective antisense oligonucleotides. Bioinformatics, 17, 1058–1061. 18. Scherr, M., Rossi, J. J., Sczakiel, G. & Patzel, V. (2000). RNA accessibility prediction: a theoretical approach is consistent with experimental studies in cell extracts. Nucl. Acids Res. 28, 2455–2461. 19. Wu¨nsche, W. & Sczakiel, G. (2005). The biological activity of siRNA is related to the kinetics of siRNAtarget RNA recognition in vitro: mechanistic implications. J. Mol. Biol., 345 203–209. 20. Matveeva, O., Felden, B., Tsodikov, A., Johnston, J.,

Target Structure and siRNA Efficacy

21. 22. 23.

24.

25. 26.

Monia, B. P., Atkins, J. F. et al. (1998). Prediction of antisense oligonucleotide efficacy by in vitro methods. Nature Biotechnol. 16, 1374–1375. Kurreck, J. (2003). Antisense technologies. Improvement through novel chemical modifications. Eur. J. Biochem. 270, 1628–1644. Sczakiel, G. (2002). Theoretical and experimental approaches to design effective antisense oligonucleotides. Front. Biosci., 5, 194–201. Grunweller, A., Wyszko, E., Bieber, B., Jahnel, R., Erdmann, V. A. & Kurreck, J. (2003). Comparison of different antisense strategies in mammalian cells using locked nucleic acids, 2 0 -O-methyl RNA, phosphorothioates and small interfering RNA. Nucl. Acids Res. 31, 3185–3193. Westerhout, E. M., Ooms, M., Vink, M., Das, A. T. & Berkhout, B. (2005). HIV-1 can escape from RNA interference by evolving an alternative structure in its RNA genome. Nucl. Acids Res. 33, 796–804. Zuker, M. (1989). On finding all suboptimal foldings of an RNA molecule. Science, 244, 48–52. Mathews, D. H., Sabina, J., Zuker, M. & Turner, D. H. (1999). Expanded sequence dependence of thermodynamic parameters improves prediction of RNA secondary structure. J. Mol. Biol. 288, 911–940.

881 27. Rychlik, W. & Rhoads, R. E. (1989). A computer program for choosing optimal oligonucleotides for filter hybridization, sequencing and in vitro amplification of DNA. Nucl. Acids Res. 21, 8543–8551. 28. Ho, S. P., Bao, Y., Lesher, T., Malhotra, R., Ma, L. Y., Fluharty, S. J. & Sakai, R. R. (1998). Mapping of RNA accessible sites for antisense experiments with oligonucleotide libraries. Nature Biotechnol. 16, 59–63. 29. Takahashi, K., Sawasaki, Y., Hata, J., Mukai, K. & Goto, T. (1990). Spontaneous transformation and immortalization of human endothelial cells. In Vitro Cell Dev. Biol. Anim. 26, 265–274. 30. Dirks, W. G., MacLeod, R. A. & Drexler, H. G. (1999). ECV304 (endothelial) is really T24 (bladder carcinoma): cell line cross-contamination at source. In Vitro Cell Dev. Biol. 35, 558–559. 31. Schubert, S., Gru¨nweller, A., Erdmann, V. A. & Kurreck, J. (2005). Local RNA target structure influences SiRNA efficacy: systematic analysis of intentionally designed binding regions. J. Mol. Biol. (this issue).

Edited by J. Karn (Received 9 December 2004; received in revised form 2 March 2005; accepted 2 March 2005)