The efficacy of small interfering RNAs targeted to the type 1 insulin-like growth factor receptor (IGF1R) is influenced by secondary structure in the IGF1R transcript.

The type 1 insulin-like growth factor receptor (IGF1R) is often overexpressed by tumors and mediates growth and apoptosis protection. We previously showed that antisense reagents complementary to the IGF1R translation start site enhance radio- and chemosensitivity and impair Atm function. However these agents induce relatively modest IGF1R down-regulation and affect insulin receptor levels. To identify alternative sites for molecular targeting, we utilized scanning oligonucleotide arrays to probe the secondary structure of IGF1R mRNA. This strategy enabled selection of antisense oligonucleotides that generated high heteroduplex yield with IGF1R but not insulin receptor transcripts. Antisense oligonucleotides that hybridized strongly to IGF1R mRNA caused IGF1R down-regulation within intact tumor cells, whereas weakly hybridizing oligonucleotides were inactive. Furthermore, the ability of small interfering RNAs (siRNAs) to block IGF1R expression correlated with the accessibility of the target sequence within the transcript. Thus, siRNAs corresponding to weakly hybridizing oligonucleotides caused minor IGF1R down-regulation, whereas siRNAs homologous to accessible targets induced profound sequence-specific IGF1R gene silencing, blocked IGF signaling, and enhanced tumor cell radiosensitivity. This indicates that secondary structure in the target transcript has a major effect on siRNA efficacy. These findings have implications for siRNA design and suggest that IGF1R-targeting agents incorporating this mode of action have potential as anticancer therapy.

The type 1 insulin-like growth factor receptor (IGF1R) 1 is often overexpressed by tumors (1)(2)(3), and IGF1R activation mediates tumor cell proliferation, motility, and protection from apoptosis (4). Tumor growth can be inhibited in vivo by blocking IGF1R expression using antisense agents targeting the IGF1R translation start site (TSS) (5). We previously showed that TSS antisense oligonucleotides (ASOs) and antisense RNA enhanced tumor cell sensitivity to cytotoxic drugs and ionizing radiation and impaired the function of Atm (6,7). However we could not suppress IGF1R expression by more than 80% and sought to identify alternative sites for molecular targeting.
Intramolecular folding of mRNAs renders all but 5-10% of most transcripts inaccessible to binding of complementary nucleic acids, but the complex secondary structure of long mRNAs is not amenable to accurate modeling (8 -11). We used an array-based screen (12) to identify sites within the human IGF1R transcript that were accessible to RNase-H-mediated cleavage. This information on secondary structure allowed us to identify molecular agents that induced IGF1R down-regulation without affecting insulin receptor (IR) expression. Structural constraints were shown to govern the activity of ASOs and also of small interfering RNAs (siRNAs) that mediate RNA interference (RNAi) in mammalian cells (13).

MATERIALS AND METHODS
Oligonucleotides-A 12-mer deoxyribophosphodiester oligonucleotide library (dN 12 ) was synthesized as described (12). Phosphorothioate oligonucleotides were synthesized and HPLC-purified at the Cancer Research UK Oligonucleotide Laboratory, South Mimms, UK. RNA/ DNA chimeric oligonucleotides were designed as described (13); the 5Ј base of each sense strand was immediately downstream of an AA doublet, and each strand incorporated 19 bases of RNA with two 3Ј deoxythymidines. An inverted sequence control duplex was made for each siRNA. These oligonucleotides were synthesized and HPLC-purified at Transgenomic Laboratories, Glasgow UK. Complementary strands were annealed in 100 mM potassium acetate, 30 mM Hepes-KOH, pH 7.4, 2 mM magnesium acetate by incubating at 85°C for 1 min followed by 37°C for 1 h. Duplex formation was confirmed by electrophoresis through 5% low melting temperature agarose (NuSieve GTG, FMC BioProducts, Rockland ME).
RNase Mapping-The 1.6-kilobase HindIII-Asp718 fragment of human IGF1R cDNA was cloned into HindIII and Asp718-digested pBS KS-(Stratagene). This plasmid was digested with NotI, blunt-ended with Klenow enzyme, digested with EcoRV, and re-ligated to remove extraneous sequence between the T7 promoter and 5Ј end of the insert. The resulting construct pIGF1R1581 ⌬55 was linearized with Asp718. End-labeled transcripts representing 1-1581 nt of IGF1R mRNA were made by in vitro transcription at 30 or 37°C with 50 Ci of [␥-32 P]GTP (Ͼ5000 Ci/mmol, Amersham Biosciences), 750 M each ATP, CTP, and UTP, and 75 M unlabeled GTP. Transcripts were purified using MicroSpin G-25 columns (Amersham Biosciences) and analyzed on 6% denaturing polyacrylamide gels. For RNase mapping reactions, 5 fmol of 5Ј end-labeled transcript were incubated with 0.5 units of RNase H and 0 -500 pmol of deoxyribophosphodiester oligonucleotide library (dN 12 ), and the products were analyzed by gel electrophoresis.
Antisense Oligonucleotide Scanning Arrays-Arrays were made as described (12) using amidated polypropylene (Beckman Instruments, Fullerton, CA) as substrate. The synthesis was carried out using standard nucleotide-cyanoethyl-phosphoramidites on an adapted ABI 394 DNA synthesizer (Applied Biosystems). Reagents were delivered to the substrate surface using a diamond-shaped mask with a 30-mm diagonal. The mask was sealed against the polypropylene to create a cell into which reagents for first base synthesis were delivered and then removed. After each nucleotide coupling the mask was displaced by 1. After 30 min at 37°C reactions were terminated by the addition of 10 l of formamide gel loading buffer, and the products were analyzed by electrophoresis through 6% denaturing polyacrylamide gels.
Cell Culture and Transfection-Tumor cell lines MDA-MB-231 (human estrogen receptor negative breast cancer), ME (human melanoma (15), A549 (human non-small cell lung cancer), DU145 (human androgen-resistant prostate cancer), UC101 (human ovarian cancer), and B16.F1 (mouse melanoma) were cultured in RPMI 1640 medium with 10% fetal calf serum, and all were negative for mycoplasma infection. Cells were transfected with ASOs using Cytofectin (Glen GSV) or Oligofectamine (Invitrogen) according to the manufacturer's instructions. All siRNA transfections were performed using Oligofectamine. Total RNA was made using RNeasy columns (Qiagen) and 0.5 g of each sample was analyzed for expression of IGF1R and actin using the Access reverse transcription-PCR kit (Promega) according to the manufacturer's instructions. The sequences of PCR primers were: IGF1R forward, 5Ј-gaaatctgcgggccaggcatcg-3Ј; IGF1R reverse, 5Ј-ctccatggtccctggacacaggt-3Ј; actin forward, 5Ј-tcatgaagtgtgacgttgacatccgt-3Ј; actin reverse, 5Ј-cctagaagcatttgcgttgcacgatg-3Ј. PCR cycles were 48°C for 45 min, 94°C for 2 min, followed by 20 cycles (shown in preliminary reactions to give quantitative yield of product) of 94°C for 30 s, 54°C for 45 s, 68°C for 2 min, with a final 7 min at 68°C. After 48 h some transfected cultures were serum-starved overnight and stimulated with 50 nM long-R3-IGF-I (GroPep). Cells were lysed and analyzed as described (6) by immunoblotting for IGF1R (Santa-Cruz), IR (Santa Cruz), phospho-Ser-473 Akt (Cell Signaling Technology), total Akt (Cell Signaling Technology), and ␤-tubulin (Sigma). Immunofluorescent detection of the IGF1R used antibody ␣IR3 (Ab1, Oncogene Science) as Numbered arrowheads, ASO sequences selected for further study. The 5Ј end of the sense sequence (17) targeted by ASO1 was base 596, that targeted by ASO2 was base 612, that targeted by ASO3 was base 622, that targeted by ASO4 was base 637, and that targeted by ASO6 was base 546. C, left, 5Ј end-labeled IGF1R transcript was incubated with RNase H in the absence of ASO or with ASOs 1, 2, 6 or TSS ASO at increasing molar excesses of ASO:transcript (1:1, 5:1, 20:1, 30:1, 100:1). Arrowheads, full-length ϳ1600-nt transcript, 600-nt cleavage product of ASOs 1 and 2, 50-nt product of TSS ASO. Asterisk, nonspecific product present in all lanes. The lower part of the gel is shown at longer exposure to visualize the 50-nt product. Right, analysis (mean Ϯ S.E.) of 3 assays, expressed as % 600-nt cleavage product. A greater yield of product is shown in the presence of 20:1 to 100:1 ASOs 1 and 2 than ASO6, and ASO2 was more effective than ASO1 at 5:1 and 20:1 (p Ͻ 0.05 for each comparison).
described (16). Activated IGF1R was detected by immunoprecipitation for phosphotyrosine (P-Tyr-100, Cell Signaling Technology) and immunoblotting for IGF1R ␤-subunit. To measure survival, cells were disaggregated 24 h after transfection and re-seeded in 6-cm dishes at 2000 cells/dish. After 4 -6 h to allow adherence, some dishes were irradiated in a 137 Cs source at 3 gray/min. After 1-2 weeks of incubation at 37°C, visible colonies were stained and counted (7). Statistical analysis was performed with GraphPad Prism 3.0c software using the paired t test for comparisons of two treatments and analysis of variance for multiple comparisons.

RESULTS
We used a two-step empirical screen (12,14) to investigate binding accessibility in the IGF1R transcript and to identify structurally homologous regions within the insulin receptor transcript. First, RNase mapping identified broad regions of accessibility within the 5Ј 1.6 kilobases of the IGF1R transcript (Fig. 1A). Based on the relative yield of cleavage products and the mixed nature of the sequence, we selected for further study a 150-nt region corresponding to bases 536 -685 of the human IGF1R sequence (17). To locate more precisely the regions of high and low accessibility, we synthesized ASO-scanning arrays complementary to this region. The ASOs were covalently bound by their 3Ј ends to the aminated polypropylene substrate and represented all possible complementary sequences of 1-20 residues within the selected region. Probing with IGF1R mRNA at 37°C (Fig. 1B) showed negligible binding to monomers at the edges of the array. Along the center line 20-mers showed significant heteroduplex yield only in the accessible regions of transcript, with three peaks of hybridization in the region 590 -668 nucleotides (peaks 1, 2, 4; Fig. 1B). Hybridizations at 25°C resulted in higher heteroduplex yield (not shown), with a similar hybridization pattern and an additional focus of hybridization not present at 37°C (region 3, Fig. 1B). We also hybridized the array to human IR mRNA to identify ASOs that could affect expression of the IR, which has 60% sequence homology with the IGF1R (Fig. 1B). This analysis revealed two peaks of heteroduplex formation between IR mRNA and the IGF1R ASOs on the array (lower panel, Fig. 1B).
We synthesized 20-mer phosphorothioate ASOs that showed intense (ASOs 1, 2, 4) or negligible (ASOs 3, 6) binding to IGF1R mRNA at 37°C (Fig. 1B and Table I). In a cell-free assay, these ASOs induced RNase H-mediated cleavage of IGF1R mRNA that was proportional to the relative affinity of hybridization when tethered on the scanning array (Fig. 1C). Effects on IGF1R expression in human tumor cells were assessed using transfection conditions that resulted in uptake of fluorescently tagged ASO into 90 -95% of cells 5-24 h after transfection (not shown). We observed significant IGF1R downregulation in MDA-MB-231 breast cancer cells transfected with 300 nM ASOs 1 (p Ͻ 0.01), 2 (p Ͻ 0.01), and 4 (p Ͻ 0.05) that hybridized strongly to the transcript at 37°C (Fig. 2, A and B). There was no effect on IGF1R expression in cells transfected with ASOs 3 and 6 that hybridized weakly at 37°C (Fig. 2, A  and B). The same patterns of relative efficacy were observed in ME human melanoma and B16.F1 murine melanoma (not shown).
We compared the novel ASOs with an IGF1R ASO complementary to the IGF1R TSS, the conventional site for antisense targeting (5). In cell-free RNase H assays, this sequence generated an indistinct 50-nt cleavage product (Fig. 1C). In transfected MDA-MB-231 cells ASOs 1, 2, and 4 caused significantly greater IGF1R down-regulation than the TSS ASO at 30 nM (p Ͻ 0.01 for each comparison), but there was no difference at 300 nM (Fig. 2B). In addition we saw evidence of modest but consistent down-regulation of the insulin receptor by 300 nM TSS ASO (69 Ϯ 8% of levels in scrambled control-treated cells compared with 112 Ϯ 11% for ASO4, p Ͻ 0.01; Fig. 2C). We did not investigate secondary structure in the region of the IGF1R translation start site, and the TSS ASO has negligible homology to the start site of the IR transcript. However several regions of IR mRNA have 4 -5 bases of homology with the TSS ASO, which may be sufficient (if accessible to binding) to induce RNase H activity (18). The array screen had revealed two regions of heteroduplex formation between IR mRNA and the IGF1R ASO array, one immediately upstream of ASO1 (Fig.  1B). This ASO hybridized with ϳ30-fold greater intensity to the IGF1R transcript than to the IR and caused minimal effects on IR expression (85-90% of control levels at 150 -300 nM). There was no detectable IR down-regulation in cells transfected with ASOs 2 or 4 ( Fig. 2C and data not shown). Thus, the scanning array permitted identification of ASOs that induced sequencespecific IGF1R down-regulation without affecting expression of the IR.
This study was extended to evaluate the effects of secondary structure on IGF1R down-regulation induced by siRNAs. We synthesized a 21-mer RNA duplex R2 corresponding to ASO2, which was immediately downstream of the required AA motif (13), and that was the most intensely hybridizing ASO from the scanning array (Fig. 1B). Duplex R2 induced profound sequence-specific IGF1R gene silencing, to ϳ1% of levels in cells treated with an inverted control duplex (Fig. 3). This suggests that the inability to suppress IGF1R expression by greater than 80% using antisense was unlikely to be due to poor transfection efficiency but was more likely a fundamental limitation of the AS approach. An 18-mer R2 duplex with a 3-bp 3Ј deletion was less effective than the 21-bp R2, whereas comparable IGF1R down-regulation was induced by 24-and 27-mer duplexes representing 3-and 6-bp 3Ј extensions of R2 (see Table I and Fig. 3D). This is in contrast to duplex length requirements for RNAi in Drosophila (19) but consistent with recently reported characteristics of RNAi in mammalian cytoplasmic lysate (20).
The effects of R2 were compared at the RNA and protein levels with R6, a duplex corresponding to ASO6 that failed to hybridize to IGF1R mRNA on the array (Fig. 1B). Compared with R6, duplex R2 caused greater inhibition of expression of the IGF1R measured by reverse transcription-PCR (Fig. 4A) and induced more profound dose-dependent reduction in IGF1R protein levels than R6 (p Ͻ 0.05 at 0.5 and 50 nM, p Ͻ 0.01 at 5 nM; Fig. 4, B and C). Indeed the difference between these two duplexes was greater than that between R2 and a mutant R2 duplex (Fig. 4, B and C). It was notable that in this system a single base pair mutation reduced but did not abolish siRNA activity. R2 and R6 were also compared in a range of human and murine cell lines, and in all cases R2 caused significantly more profound IGF1R down-regulation (DU145 (p Ͻ 0.01), A549 (p Ͻ 0.05), ME (p Ͻ 0.01), B16 (p Ͻ 0.05); Fig. 4D). This indicates that secondary structure in the transcript has a major effect on siRNA efficacy and suggests that the structural features dictating access are robust and conserved between different cell lines and species.
We then assessed the efficacy of a second pair of duplexes based on the sequence around peak 4 of hybridization between IGF1R mRNA and the IGF1R ASO scanning array (Fig. 5A). Duplexes R4 and R5 were designed to target 19-mer sequences immediately downstream of AA motifs at bases 636 and 639, respectively, of the IGF1R sequence (17). Data from the scanning array indicated that there was a ϳ6.5-fold difference in hybridization intensity between the equivalent 19mer ASOs (Fig. 5B). Both duplexes induced IGF1R down-regulation in human ovarian and prostate cancer cells, but R4 was significantly more potent (p Ͻ 0.05) in both cell lines (Fig. 5, C and D). Thus, a 3-bp shift had major effects on siRNA efficacy, paralleling differences in heteroduplex yield on the scanning array.
Next we investigated the effects of ASOs and siRNA on intracellular signaling. The relatively modest IGF1R downregulation induced by ASOs 1-6 or siRNA R6 was not sufficient to abolish IGF-I signaling to Akt (Fig. 6A and data not shown). In contrast transfection with R2 induced profound IGF1R gene silencing such that we could not detect IGF-I-induced phospho-rylation of the IGF1R ␤-subunit or of Akt, a key intermediate in the major anti-apoptosis pathway downstream of the IGF1R (4). Finally we compared the effects on colony-forming efficiency and post-irradiation survival of transfection with 100 nM ASO4 or siRNA duplexes R4 and R5, equivalent to strongly and weakly hybridizing ASOs, respectively. The phosphorothioate ASO4 caused inhibition of overall survival and survival after 2 gray irradiation (SF2), but this effect was entirely nonspecific, with similar results in cultures treated with the scrambled sequence oligonucleotide (Scr4, Fig. 6, B and C). Transfection with siRNAs R4 or R5 inhibited survival of MDA-MB-231 (p Ͻ 0.001 for comparison of each siRNA with inverted control duplex), with a significantly greater effect in R4-treated cells (p Ͻ 0.01 for comparison with R5; Fig. 6B). Only in cultures treated with the more potent siRNA duplex R4 was there significant FIG. 3. Profound IGF1R gene silencing induced by siRNA equivalent to IGF1R ASO that hybridizes intensely to IGF1R mRNA. A, human ME melanoma cells were transfected with Oligofectamine alone (Nil) or with 200 nM TSS scrambled sequence control (Scr) or TSS ASO or with siRNA R2 or inverted sequence duplex (Inv2). Cells were lysed 48 h after transfection and analyzed by immunoblotting for IGF1R. Arrow, ϳ220-kDa IGF1R proreceptor. B, sequence specificity of IGF1R gene silencing in ME melanoma cells transfected with Oligofectamine alone or with 5, 50, or 500 nM inverted sequence duplex or R2. Cell lysates were analyzed by immunoblotting for IGF1R and IR. Similar results were obtained in a second set of independently prepared ME cell lysates, in MDA-MB-231 breast cancer and UC101 ovarian cancer cells. C, immunofluorescent staining of MDA-MB-231 cells 48 h after transfection with 100 nM inverted sequence duplex or R2. Red, IGF1R; blue, Hoescht stain for DNA. D, MDA-MB-231 cells were transfected with 0.5 or 5 nM R2 or with 18 -27-mer variants of the same duplex (see Table I). IGF1R protein levels were analyzed after 48 h. Arrow, ϳ220-kDa IGF1R proreceptor. inhibition of survival after 2 gray irradiation (SF2; p Ͻ 0.01 for comparisons with Inv4 and R5; Fig. 6C). These results confirm that the complex secondary structure of target transcripts can influence siRNA efficacy and that these differences are sufficient to confer major variation in biological efficacy of potential therapeutic significance. DISCUSSION We previously showed correlation between array hybridization and effects on gene expression, by assessing in vitro translation of cyclin B transcripts following ASO microinjection into Xenopus oocytes (12). Here, we have shown that intensity of array hybridization at 37°C successfully predicted for target down-regulation within intact tumor cells. This is despite the fact that array conditions were non-physiological; ASOs were tethered at the 3Ј ends, and the composition of the hybridization buffer did not reflect the protein-rich cytosolic microenvironment. Furthermore, hybridizations were performed in 1 M NaCl since there was insignificant heteroduplex formation at physiological (150 mM) salt concentration (not shown). In a search for effective IGF1R ASOs to combat epithelial hyperplasia in psoriasis, Wraight et al. (21) used a program designed to predict regions of the transcript lacking internal duplex and hairpin structure (21). One of the selected 15-mer ASOs fell within the region we screened, targeting nt 653-667 of the IGF1R transcript. This was found to be ineffective and indeed was shown to hybridize weakly on our scanning array (relative hybridization intensity ϳ0.03 compared with 1.0 for ASO2; Fig.  1B). In accordance with our results, even the most effective ASOs suppressed IGF1R expression only to ϳ25% of control levels (21).
We found major differences in the activity of IGF1R siRNAs, paralleled by changes in the intensity of hybridization of the equivalent ASOs to IGF1R mRNA. There have been two recent reports of differential efficacy of siRNAs, against human immunodeficiency virus-1 rev (22) and Tissue Factor (23), attributed in the latter report to protein binding at/near the siRNA target site. This is an unlikely explanation for our findings since we observed parallel differences in ASO and siRNA efficacy within intact cells and in cell-free (and virtually proteinfree) RNase H assays. The parallels we observed between array hybridization and siRNA efficacy suggest that secondary structure in the IGF1R transcript was a major factor in determining the efficacy of synthetic 21-mer siRNAs. Unlike phosphorothioate ASOs that induce RNase H activity, siRNAs act as "guide sequences" to direct the RNA-induced silencing complex to the homologous sequence in the target transcript (24). The precise nature of the molecular interaction between siRNAs and target mRNA is unclear. The need for access comparable with that required for ASO binding supports the concept, as originally proposed when RNAi was first recognized (24,25), of direct interaction by base-pairing between the transcript and component(s) of the duplex. This is consistent with the recent demonstration that antisense strands can mediate RNAi in mammalian cytoplasmic lysate (20).
The success rate for effective siRNA design is clearly higher than the ϳ10% reported for ASOs (10,26). However a requirement for access could explain why some siRNAs are ineffective. Elbashir et al. (13) were unable to down-regulate vimentin in HeLa cells, but in a later report vimentin expression was effectively silenced by siRNAs directed against different regions of the same transcript (26). We have demonstrated here that mRNA folding has major consequences for siRNA-induced gene silencing, with clear implications for the design of effective siRNAs. Furthermore, the potency of these siRNAs in blocking IGF signaling and tumor cell survival suggests that IGF1Rtargeting agents incorporating this mode of action have potential as anticancer therapy.