Efficient Reduction of Target RNAs by Small Interfering RNA and RNase H-dependent Antisense Agents

RNA interference can be considered as an antisense mechanism of action that utilizes a double-stranded RNase to promote hydrolysis of the target RNA. We have performed a comparative study of optimized antisense oligonucleotides designed to work by an RNA interference mechanism to oligonucleotides designed to work by an RNase H-dependent mechanism in human cells. The potency, maximal effectiveness, duration of action, and sequence specificity of optimized RNase H-dependent oligonucleotides and small interfering RNA (siRNA) oligonucleotide duplexes were evaluated and found to be comparable. Effects of base mismatches on activity were determined to be position-dependent for both siRNA oligonucleotides and RNase H-dependent oligonucleotides. In addition, we determined that the activity of both siRNA oligonucleotides and RNase H-dependent oligonucleotides is affected by the secondary structure of the target mRNA. To determine whether positions on target RNA identified as being susceptible for RNase H-mediated degradation would be coincident with siRNA target sites, we evaluated the effectiveness of siRNAs designed to bind the same position on the target mRNA as RNase H-dependent oligonucleotides. Examination of 80 siRNA oligonucleotide duplexes designed to bind to RNA from four distinct human genes revealed that, in general, activity correlated with the activity to RNase H-dependent oligonucleotides designed to the same site, although some exceptions were noted. The one major difference between the two strategies is that RNase H-dependent oligonucleotides were determined to be active when directed against targets in the pre-mRNA, whereas siRNAs were not. These results demonstrate that siRNA oligonucleotide- and RNase H-dependent antisense strategies are both valid strategies for evaluating function of genes in cell-based assays.

RNA interference has become a powerful and widely used tool for the analysis of gene function in invertebrates and plants (1,2). Introduction of long double-stranded RNA into the cells of these organisms leads to the sequence-specific degradation of homologous gene transcripts. The long doublestranded RNA molecules are metabolized to small 21-23-nu-cleotide interfering RNAs (siRNAs) 1 by the action of an endogenous ribonuclease, Dicer (3,4). The siRNA molecules bind to a protein complex, termed RNA-induced silencing complex, which contains a helicase activity that unwinds the two strands of RNA molecules, allowing the antisense strand to bind to the targeted RNA molecule (4,5) and an endonuclease activity that hydrolyzes the target RNA at the site where the antisense strand is bound. It is unknown whether the antisense RNA molecule is also hydrolyzed or recycles and binds to another RNA molecule. Therefore, RNA interference is an antisense mechanism of action, since ultimately a single-stranded RNA molecule binds to the target RNA molecule by Watson-Crick base pairing rules and recruits a ribonuclease that degrades the target RNA.
In mammalian cells, long double-stranded RNA molecules were found to promote a global change in gene expression, obscuring any gene-specific silencing (6,7). This reduction in global gene expression is thought to be mediated in part through activation of double-stranded RNA-activated protein kinase which phosphorylates and inactivates the translation factor eukaryotic initiation factor 2␣ (8). Recently, it has been shown that transfection of synthetic 21-nucleotide siRNA duplexes into mammalian cells does not elicit the RNA-activated protein kinase response, allowing effective inhibition of endogenous genes in a sequence-specific manner (9,10). These siR-NAs are too short to trigger the nonspecific double-stranded RNA responses, but they still promote degradation of complementary RNA sequences (9,11).
Multiple mechanisms exist by which short synthetic oligonucleotides can be used to modulate gene expression in mammalian cells (12). A commonly exploited antisense mechanism is RNase H-dependent degradation of the targeted RNA. RNase H is a ubiquitously expressed endonuclease that recognizes a DNA-RNA heteroduplex, hydrolyzing the RNA strand (13,14). Antisense oligonucleotides that contain at least five consecutive deoxynucleotides are substrates for human RNase H (15,16). Thus, the RNase H-dependent antisense mechanism differs from the siRNA mechanism by utilizing RNase H, instead of a double-stranded RNase, as the terminating mechanism.
Initial reports in which siRNA was compared with singlestranded antisense approaches to gene knockdown have indicated that the siRNA is more potent and effective than a traditional antisense approach (4,10). However, the antisense molecules used in these experiments were single-stranded unmodified RNA, which is rapidly degraded and does not recruit RNase H to cleave the target. Phosphorothioate oligodeoxynucleotides are first generation antisense agents that have been widely used to modulate gene expression in cell-based assays, in animal models, and in the clinic (18). The phosphorothioate modification dramatically increases the nuclease resistance of the oligonucleotide and still supports RNase H activity (19). Further improvements to phosphorothioate oligodeoxynucleotides have been made, resulting in second generation oligonucleotides such as 2Ј-O-methyl or 2Ј-O-methoxyethyl modifications (15,20). The 2Ј-O-methoxyethyl modification is particularly attractive, since it increases the potency of the oligonucleotide, further increases nuclease resistance, decreases toxicity, and increases oral bioavailability (21)(22)(23)(24).
In this report, we compare oligonucleotides that were designed to work by a siRNA mechanism (siRNA oligonucleotides) to optimized first and second generation antisense oligonucleotides that were designed to work by an RNase Hdependent mechanism (RNase H oligonucleotides). Active siRNA oligonucleotides and homologous RNase H-dependent oligonucleotides were evaluated for relative potency, efficacy, duration of action, sequence specificity, and site of action within the cell to determine whether significant advantages could be found for the different antisense strategies in cellbased assays. Our results suggest that in human cell culture assays, double-stranded oligoribonucleotides that work by siRNA mechanism exhibit similar potency, efficacy, specificity, and duration of action as RNase H oligonucleotides. Furthermore, as we have previously found for RNase H oligonucleotides, not all sites on the target RNA are good target sites for siRNA molecules. Like RNase H-dependent oligonucleotides, activity of siRNAs is affected by the secondary structure of the target RNA. Finally, siRNAs and RNase H oligonucleotides appear to work upon the target mRNA at different stages of its processing/metabolism.

EXPERIMENTAL PROCEDURES
Oligonucleotide Synthesis and siRNA Duplex Formation-Synthesis and purification of phosphorothioate-modified oligodeoxynucleotides or chimeric 2Ј-O-methoxyethyl/deoxyphosphorothioate modified oligonucleotides was performed using an Applied Biosystems 380B automated DNA synthesizer as described previously (22). Sequences of oligonucleotides and placement of 2Ј-O-methoxyethyl modifications are detailed in Tables I and II. RNA oligonucleotides were synthesized at Dharmacon Research, Inc. (Boulder, CO). siRNA duplexes were formed by combining 30 l of each 50 M RNA oligonucleotide solution and 15 l of 5ϫ annealing buffer (100 mM potassium acetate, 30 mM HEPES-KOH, pH 7.4, 2 mM magnesium acetate) followed by heating for 1 min at 90°C and then 1 h at 37°C. Successful annealing was confirmed by nondenaturing polyacrylamide gel electrophoresis. The melting temperatures (T m ) were experimentally determined for a subset of siRNA tested as described previously (15). In each case, the measured T m values were greater than 55°C. The predicted T m values for all siRNA duplexes used in this paper were Ͼ50°C (100 mM salt, 0.1 M oligonucleotide).
Cell Culture-T24 cells, (American Type Tissue Culture Collection, Manassas, VA) were cultivated in Dulbecco's modified Eagle's medium supplemented with 10% fetal bovine serum in six-well culture dishes at a density of 250,00 cells/well. Oligonucleotides were administered to cells using Lipofectin reagent (Invitrogen) as described previously (25,26). Other transfection reagents were evaluated (e.g. Transit TKO, LipofectAMINE 2000, and Oligofectamine) and found to provide similar levels of siRNA-mediated target reduction in T24 cells (data not shown); however, Lipofectin was determined to be superior to the other transfection reagents for RNase H-dependent oligonucleotide administration. In addition, LipofectAMINE was found to be more toxic to the cells than Lipofectin, and Transit TKO failed to provide consistent results for delivery of siRNA molecules. Optimal Lipofectin/oligonucleotide ratios were empirically determined for both siRNAs and RNase H-dependent oligonucleotides. For RNase H antisense oligonucleotides, cells were incubated with a mixture of 3 g/ml Lipofectin per 100 nM oligonucleo-tide in OptiMEM medium (Invitrogen), whereas siRNA duplexes were incubated with a mixture of 6 g/ml Lipofectin per 100 nM siRNA duplex. Since concentrations reported in the paper represent concentration of the siRNA duplex, the same weight/Lipofectin ratio was maintained for siRNA duplexes and antisense oligonucleotides. After 4 h, the transfection mixture was aspirated from the cells and replaced with fresh Dulbecco's modified Eagle's medium plus 10% fetal calf serum and incubated at 37°C, 5% CO 2 until harvest.
To induce CD54 mRNA expression, oligonucleotide-treated cells were incubated overnight and then treated with 5 ng/ml TNF-␣ (R&D Systems, Minneapolis, MN) for 2-3 h prior to harvest of cells for RNA expression analysis. For analysis of cell surface expression of CD54 protein, cells were induced with 5 ng/ml TNF-␣ immediately following the transfection and incubated overnight.
Flow Cytometry-Following oligonucleotide treatment, cells were detached from the plates with Dulbecco's phosphate-buffered saline (without calcium and magnesium) supplemented with 4 mM EDTA. Cells were transferred to microcentrifuge tubes, pelleted at 5000 rpm for 1 min, and washed in 2% bovine serum albumin, 0.2% sodium azide in Dulbecco's phosphate-buffered saline at 4°C. PE anti-human CD54 antibody (catalog no. 555511; Pharmingen, San Diego, CA) was then added at 1:20 in 0.1 ml of the above buffer. The antibody was incubated with the cells for 30 min at 4°C in the dark. Cells were washed again as above and resuspended in 0.3 ml of PBS buffer with 0.5% paraformaldehyde. Cells were analyzed on a Becton Dickinson FACScan. Results are expressed as percentage of control expression based upon the mean fluorescence intensity.
Luciferase Assays-For luciferase-based reporter gene assays, 10 g of plasmid pGL3-5132-S0 or pGL3-5132-S20 (26) was introduced into COS-7 cells at 70% confluence in 10-cm dishes using SuperFect Reagent (Qiagen). Following a 2-h treatment, cells were trypsinized and split into 24-well plates. Cells were allowed to adhere for 1 h, and then RNase H or siRNA oligonucleotides were added in the presence of Lipofectin reagent as detailed above. All oligonucleotide treatments were performed in duplicate or triplicate. Following the 4-h oligonucleo-tide treatment, cells were washed, and fresh Dulbecco's modified Eagle's medium containing 10% fetal calf serum was added. The cells were incubated overnight at 37°C. The following morning, cells were harvested in 150 l of passive lysis buffer (Promega, Madison, WI), and 60 l of lysate was added to each well of a black 96-well plate followed by 50 l of luciferase assay reagent (Promega). Luminescence was measured using a Packard TopCount microplate scintillation counter.
Statistical Analyses of Gene Screen Data-Simple statistical analyses were conducted to examine the association between siRNA and RNase H oligonucleotide screens. Similarity between the two screens for a given gene was measured by using correlation coefficients and average difference. Two different correlation measures were employed: Pearson's product-moment correlation coefficient, which measures a linear relationship between siRNA and RNase H oligonucleotide screens, and Spearman's rank-order correlation coefficient, which measures a linear relationship between the potency of siRNA and RNase H oligonucleotide screens. One-sample one-tailed t tests were conducted for observed correlation coefficients to assess whether they are significantly greater than the null hypothesis of no correlation. Statistical inference on observed average difference was conducted by randomizing sample pairs of siRNA and RNase H oligonucleotide screen. Again, one-tailed tests were used to determine whether the observed distances are significantly smaller than those expected from random chance. The association between siRNA and RNase H oligonucleotide screen was further examined by the receiver operating characteristic (ROC) analysis. First, siRNAs were classified as potent when the percentage inhibition rate was smaller than the median value of 67.4% for the CD54 siRNA screen and 57.1% for the PTEN screen. An arbitrary cut-off was then set for RNase H oligonucleotide screens. RNase H oligonucleotides with percentage inhibition rates smaller than this cut-off value were classified as potent. From the classification of siRNAs and RNase H oligonucleotides, a 2 ϫ 2 contingency table was constructed. Finally, true positive rate (TPR) and false positive rate (FPR) were determined based on this table. For example, TPR is the number of cases where potent RNase H oligonucleotides correspond to potent siRNAs divided by the number of potent siRNAs. Similarly, FPR is the number of cases where potent RNase H oligonucleotides corresponds to nonpotent siRNAs divided by the number of nonpotent siRNAs. For CD54, a cut-off value of 70% gives TPR ϭ 75% and FPR ϭ 45%. For the PTEN gene, a cut-off of 40% gives TPR ϭ 72% and FPR ϭ 44%. By varying these cut-off values, a ROC curve can be drawn on a plane spanned by FPR and TPR. The area under the ROC curve provides a measure of overall accuracy.

Active RNase H-dependent Antisense Oligonucleotide Target Sites Predict siRNA Target Sites-Since both siRNAs and
RNase H-dependent oligonucleotides must hybridize to target RNA and subsequently direct specific RNases to bind and cleave the bound RNA (15, 28), we examined whether an active RNase H oligonucleotide site would also be an active siRNA site. Initially, siRNAs were designed and synthesized based upon the target sequences of active RNase H oligonucleotides previously identified. ISIS 5132 is a 20-base phosphorothioate oligodeoxynucleotide that targets the 3Ј-untranslated region of human c-raf kinase mRNA and specifically reduces expression of both mRNA and protein (29). An siRNA duplex (si5132) composed of 21-nt sense and 21-nt antisense strands was designed using the first 19 nucleotides of the target site for ISIS 5132 in the paired region and unpaired 2-nt 3Ј-dTdT overhangs. T24 cells were treated with oligonucleotides at doses ranging from 3 to 300 nM as detailed under "Experimental Procedures." Total RNA was analyzed for expression of c-raf mRNA by quantitative RT-PCR. The results, shown in Fig. 1A, are normalized to GAPDH mRNA expression. Both ISIS 5132 (solid bars) and the corresponding siRNA to the same target site (open bars) were found to inhibit the expression of the c-raf kinase mRNA, each with an IC 50 of ϳ50 nM. siRNAs targeted to human CD54 and Bcl-X had no effect on the expression of c-raf (data not shown).
Chimeric oligonucleotides in which 2Ј-O-methoxyethyl (2Ј-MOE) substituted nucleosides flank a central unmodified 2Јoligodeoxynucleotide region that serves as substrate for RNase H region have been shown to have increased potency and duration of action as compared with phosphorothioate oligodeoxynucleotides (22). ISIS 16009 is a 20-base chimeric oligonucleotide that has previously been demonstrated to be an effective inhibitor of human Bcl-X (31). Another 20-base chimeric oligonucleotide, ISIS 116847, has been shown to effectively inhibit expression of the human PTEN gene (32). The siRNA versions, si16009 and si116847, as well as the homologous parent RNase H-dependent oligonucleotides were transfected into T24 cells at doses ranging from 10 to 200 nM. In both cases the 2Ј-MOE chimeric RNase H-dependent oligonucleotides (solid bars) were slightly more potent inhibitors of mRNA expression than the corresponding siRNA (open bars) (Fig. 1, B and C). In the case of Bcl-X, the RNase H-dependent oligonucleotide has an IC 50 of ϳ30 nM, whereas the siRNA version, si16009, has an IC 50 of ϳ100 nM. PTEN is more potently inhibited, with IC 50 values of 10 and 25 nM for the RNase H oligonucleotide and siRNA, respectively.
RNase H-dependent oligonucleotides and siRNAs were also compared for activity in T24 cells against CD54 (ICAM-1), a gene whose expression is induced by cytokine treatment. ISIS 2302, a first generation phosphorothioate oligodeoxynucleotide, hybridizes to the 3Ј-untranslated region of human CD54 (ICAM-1) and was previously shown to be a potent and specific inhibitor of CD54 expression (33). Whereas ISIS 2302 reduced ICAM-1 expression by 85%, si2302 had no inhibitory effect on message levels at concentrations as high as 200 nM as measured by quantitative RT-PCR (data not shown).
Screening for Optimized RNase H and siRNA Oligonucleotides-In order to identify potent antisense agents, many investigators design and test multiple oligonucleotides that target different sites and regions of the target mRNA (33,34). To determine if the lack of activity of the CD54 siRNA molecule was due to suboptimal siRNA design or to a blocking activity induced by TNF-␣ treatment, we designed 40 siRNA and 40 2Ј-MOE chimeric oligonucleotides to the same sites of the CD54 mRNA (Table I). The siRNA duplexes were composed of 21-nt sense and 21-nt antisense strands, paired in a manner to have a 19-nt duplex region and a 2-nt overhang at each 3Ј terminus ( Table I). The target sites included various regions of the human CD54 message including 5Ј-untranslated region (5Ј-UTR), coding region, and 3Ј-UTR. T24 cells were treated with oligonucleotides at a single concentration of 100 nM as described under "Experimental Procedures." Active sequences were identified in both the RNase H oligonucleotide and siRNA screens (Fig. 2). In the RNase H oligonucleotide screen (solid bars), 12 of 40 oligonucleotides were found to inhibit expression of CD54 mRNA by greater than 50% as compared with the untreated control, whereas the siRNA screen (open bars) identified 9 of 40 sequences as active by the same criteria. Comparison of the active target sites revealed that five of the nine active siRNA sites were also identified as active sites in the RNase H oligonucleotide screen. Similarly, the majority of sites where the RNase H-dependent oligonucleotide failed to inhibit expression, the siRNA also failed. The data also indicate that regions of greater activity or "hot spots" along the RNA transcript can be identified for both siRNA oligonucleotides and RNase H-dependent oligonucleotides. For example, homologous siRNAs and RNase H-dependent oligonucleotides both show good activity in the ϳ200 nucleotide span from base 1781 to 1971 of the 3Ј-untranslated region. These results demonstrate that the initial lack of activity for the CD54 directed siRNA molecules is not due to induction of an inhibitory factor by TNF-␣ treatment and that not all siRNA molecules designed to hybridize to an RNA transcript are effective.
Cell surface CD54 protein expression was also evaluated by flow cytometry. Comparison of mRNA reduction and protein reduction for the siRNA and RNase H-dependent oligonucleotides screens are shown in Fig. 3A. In general, the results are highly correlated with the same active targets identified by either mRNA or protein reduction. However, several oligonucleotides were identified that appear to produce a more robust reduction of protein compared with the corresponding RNA  (Fig. 3, A and B). Note, however, that CD54 RNA and protein were measured at different times following TNF-␣ induction, which may account for the discrepancies. Statistical analyses described under "Experimental Procedures" were applied to siRNA and RNase H oligonucleotide screening data for CD54 mRNA reduction. The data were composed of two independent RNase H oligonucleotide screens and five independent siRNA screens that were averaged to produce composite siRNA/RNase H-dependent oligonucleotide screens. Pearson's correlation coefficient was determined to be 0. 424   FIG. 2. CD54 antisense screen. A series of 40 chimeric oligonucleotides designed to work by an RNase H-dependent mechanism and a series of corresponding siRNAs were administered to T24 cells in the presence of Lipofectin transfection reagent. The following day, CD54 expression was induced, and RNA was harvested. CD54 mRNA expression was analyzed by quantitative RT-PCR. Results represent the percentage of induced CD54 mRNA relative to untreated control. Solid bars, RNase H oligonucleotides; open bars, siRNAs. The target site start position is the 5Ј-most nucleotide in the mRNA target. with a p value of 0.0032, and Spearman's correlation coefficient was 0.426 with a p value of 0.0039. The average difference between the two screens was 18.5% with a p value of 0.0056. These results indicate that a significant overlap exists between siRNA and RNase H oligonucleotide screens in terms of correlation coefficients and average difference. The association between siRNA and RNase H oligonucleotide activity was further analyzed using ROC analysis. The area under the ROC curve is a summary of the overall diagnostic accuracy of the test that measures the correspondence between potent siRNA and RNase H-dependent oligonucleotide sites. The area under the ROC curve is 0.75 for CD54, suggesting that a significant concordance exists between siRNA and RNase H-dependent oligonucleotide binding sites on target RNAs.
A second comparative analysis was performed using 36 2Ј-MOE chimeric oligonucleotides, 18 nucleotides in length, and a series of corresponding siRNAs (Table II) targeted to the human PTEN message. PTEN mRNA is constitutively expressed in T24 cells. Cells were treated with siRNAs or RNase H-dependent oligonucleotides as described under "Experimental Procedures." As defined by a target mRNA reduction of 50% or greater, 22 of the 36 RNase H-dependent oligonucleotides (solid bars) were identified as active (Fig. 4). In contrast, the siRNA screen (open bars) identified only 12 of 36 sites as active, defined by the same criteria. Of the 12 active siRNA oligonucleotide sites, 10 were shared as active with the RNase H-dependent oligonucleotide screen, with only 2 of the active siR-NAs not identified in the RNase H-dependent oligonucleotide screen.
The RNase H/siRNA oligonucleotide screens for PTEN were repeated three separate times. A statistical analysis of the composite data from the three experiments was performed as detailed above. Pearson's correlation coefficient was determined to be 0.425 with a p value of 0.0049, and Spearman's correlation coefficient was 0.318 with a p value of 0.0299. The average difference between the two screens was 21.3% with p value of 0.0038. These results suggest that a significant association exists between siRNA-and RNase H-dependent oligonucleotide screens in terms of Pearson's correlation coefficient and average difference. ROC analysis of these data give a value of 0.588 for PTEN. Whereas the data for the PTEN screens are not as highly significant as those for CD54, they do demonstrate a reasonable, although not perfect, correlation between siRNA and RNase H-dependent oligonucleotide binding sites.
Effect of RNA Secondary Structure on Activity-We have previously demonstrated that the secondary structure of the mRNA target strongly influences activity of RNase H-dependent oligonucleotides in cell culture (26). A luciferase reporter system was developed in which the target site for ISIS 5132 was cloned into the 5Ј-UTR of the luciferase reporter plasmid pGL3-Control. Sequence immediately adjacent to the target sequence was altered to form various RNA secondary structures that included the 5132 target sequence. These structures ranged from one in which the entire target site was sequestered in a 20-base stem closed by a UUGC tetraloop (pGL3-5132-S20) to one that had little predicted secondary structure likely to inhibit hybridization of RNase H oligonucleotide to target (pGL3-5132-S0) (26). The activities of ISIS 5132 and si5132 were compared using the pGL3-5132-S20 and pGL3-5132-S0 constructs. The reporter plasmids were transfected into COS-7 cells as detailed under "Experimental Procedures." Following the plasmid transfection, cells were seeded in 24-well plates and treated with ISIS 5132 or si5132 at doses ranging from 10 to 300 nM. Lysates from the treated cells were assayed for luciferase activity 16 h later. When directed against the message with no structure (pGL3-5132-S0), both ISIS 5132 (open circles) and si5132 (open triangles) effectively reduced luciferase expression in a dose-dependent manner with IC 50 values between 30 and 100 nM (Fig. 5), which is consistent with the observed IC 50 for endogenous message reduction. Conversely, neither the RNase H oligonucleotide (solid circles) nor siRNA (solid triangles) were found to inhibit luciferase expression when directed against the highly structured target (pGL3-5132-S20) at concentrations up to 300 nM. Therefore, the secondary structure of the target has an equally important effect on reduction of target RNA by both types of antisense oligonucleotides.
Sequence Specificity of RNase H-dependent Oligonucleotides and siRNA-The sequence fidelity of the RNA interference pathway has been evaluated to a limited extent in several hallmark systems, including C. elegans (35) and Drosophila cell extracts (28), and more recently in mammalian cell culture (9,36). Several investigators have reported that incorporation of one or two mismatches into a siRNA construct, with respect to the target RNA, is sufficient to disable RNA interference activity against the target RNA. A common attribute of each of the mismatch constructs tested thus far, however, has been location of the mismatches in the center domain of the construct. To further define the fidelity of the RNA interference pathway for perfect Watson-Crick base pair matched sequences, we tested an additional type of construct, wherein a mismatch was incorporated in each of the 5Ј-and 3Ј-terminal domains of the siRNA targeting PTEN (si116847). The same mismatches were also incorporated into ISIS 116847, an RNase H oligonucleotide. When the mismatches were placed in the center of the sequence, a complete loss of activity was observed for both siRNAs and RNase H-dependent oligonucleotides at a concentration of 100 nM (Fig. 6). In contrast to the duplex with two mismatches positioned in the center of the siRNA (gridded bars), the siRNA with mismatches in the outside domains (open bars) demonstrated only a moderate loss of activity in comparison with the perfect match construct (cross-hatched bars). The results for the RNase H-dependent oligonucleotide were similar, although the RNase H-dependent oligonucleotide containing mismatches on the ends demonstrated a greater loss of activity than was observed for the homologous siRNA (71% versus 52% control).
Comparison of Potency and Efficacy-Comparison of the relative potency of siRNAs directed to the same site on the target RNA as an optimized RNase H oligonucleotide revealed that the RNase H oligonucleotide exhibited similar or better potency as defined by IC 50 values compared with the siRNA (Fig. 1). The siRNA and RNase H-dependent oligonucleotides also exhibited a similar level of efficacy as defined by the maximal level of target RNA reduction. Since the siRNA molecules used for these analysis were not selected as the optimal siRNA molecules for the respective target based upon screening numerous siRNA sequences, we compared the most effective siRNA molecule derived from the siRNA screen ( Fig. 4) with an optimized second generation chimeric oligonucleotide to PTEN.
The different antisense agents, tested at concentrations ranging from 10 to 200 nM in T24 cells, produced a similar doseresponse curve with IC 50 values near 10 nM (Fig. 7A). Additionally, both agents reduced PTEN mRNA levels by greater than 90%.
Similarly, the most effective siRNA from the CD54 screen was compared with its corresponding second generation RNase H oligonucleotide, which showed a similar degree of efficacy in the primary screen. T24 cells were treated with either the siRNA, si121747, or the oligonucleotide, ISIS 121747, at concentrations ranging from 10 to 200 nM. As with PTEN, the FIG. 5. Inhibition of alternate structure clones by ISIS 5132/ si5132. Cells were transfected with luciferase reporter plasmids and then treated with chimeric RNase H-dependent oligonucleotide/siRNA at doses ranging from 3 to 300 nM. Luciferase expression was measured the following day. Results are the percentage of luciferase expression compared with the untreated control. Open circles/triangles, pGL5132-S0 target; solid circles/triangles, pGL5132-S20 target. Circles, ISIS 5132. Triangles, si5132. CD54 siRNA and chimeric oligonucleotide produced similar dose-response curves with IC 50 values of ϳ15 nM for the siRNA and 30 nM for the oligonucleotide for reduction of TNF-␣-induced CD54 mRNA expression. The efficacy was almost identical with maximal reduction of ϳ85% for both antisense agents.
Duration of Action-We compared the duration of action of a second generation RNase H oligonucleotide and siRNA in T24 cells using human Bcl-X as a target (Fig. 8). Cells were seeded in six-well dishes so that they would be 80 -90% confluent at the time of harvest. In T24 cells, inhibition of Bcl-X by siRNA (open bars) was found to be maximal at 24 h post-transfection and returned to normal levels by day 5. The results were similar for RNase H oligonucleotide treatment (solid bars) except that maximal activity was achieved at 8 h. In both cases, activity began to taper off between 48 and 96 h, and by 120 h, no significant inhibition of targeted message was seen with either the RNase H oligonucleotide or the siRNA.
Effects of Targeting Intron Sequences-To compare the site of activity of siRNA oligonucleotides and RNase H-dependent oligonucleotides and directly, siRNA duplexes were designed based upon several previously identified active RNase H oligonucleotide sites that target intron sequences (shown in Table  III), with the assumption that RNA transcripts containing introns would only be found in the nucleus. The target sites for COREST and PAK1 are contained completely within the introns, whereas the target sites for caspase recruitment domain 4 and Notch homolog 2 overlap the indicated intron/exon boundary with 10 nucleotides on either side (Table III). T24 cells were treated with the RNase H oligonucleotide or the corresponding siRNA at a single dose of 200 nM as described above. The results are shown in Fig. 9. In all cases, the RNase H oligonucleotides effectively reduced the targeted message (striped bars), whereas an RNase H oligonucleotide targeted to another gene, tumor necrosis factor receptor 2, had no effect on gene expression (gray bars). In contrast, the homologous siRNAs did not reduce mRNA levels for any of the four genes in which introns were targeted (open bars); nor was any nonspecific reduction observed using siRNAs targeted to tumor necrosis factor receptor 2 (cross-hatched bars). As a control, another gene, c-raf, was included, in which the target was in the exon. As previously demonstrated (Fig. 1A), the siRNA targeted to the c-raf exon did reduce message expression. These data support the hypothesis that siRNA activity is primarily cytoplasmic and therefore does not interact with pre-mRNA. DISCUSSION Multiple mechanisms exist by which synthetic oligonucleotides can be used to regulate gene expression in mammalian cells (12). To date, the most successful strategy has been to design oligonucleotides that hybridize to a target RNA by Watson-Crick base-pairing rules (i.e. antisense oligonucleotides). Once bound, antisense oligonucleotides can disable target RNAs by two broadly defined processes: disruption of RNA function by occupancy of critical sites and degradation of targeted RNA. Within these two broadly defined processes, multiple mechanisms are possible. Examples of "occupancy only" mechanisms include inhibition of translation (37), modulation of pre-mRNA splicing (38), and modulation of polyadenylation (39). In each case, the antisense oligonucleotides were found to be potent and selective regulators of gene expression.
Several endogenous enzymes can be exploited to promote targeted cleavage of RNAs in cells. One of the most widely exploited mechanisms is RNase H-mediated cleavage of targeted RNA. RNase H represents a ubiquitously expressed family of cellular enzymes that hydrolyze the RNA strand of an RNA-DNA heteroduplex. There are additional RNases present in mammalian cells that can be exploited for antisense inhibition of gene expression. As an example, we have reported that a single-stranded phosphorothioate modified RNA molecule can promote selective loss of Ha-ras in human cells (40).
Small interfering RNAs have been gaining widespread acceptance as a valuable tool for inhibiting gene expression in mammalian cells. In mammalian cells, like RNase H-dependent oligonucleotides, siRNAs bind to targeted RNA by Watson-Crick base pairing and induce site-specific cleavage of the RNA by specific RNases. The RNase that recognizes the duplex formed by the siRNA molecule has not been identified to date; however, the substrate specificity suggest that it is a doublestranded specific RNase (28). Since siRNA is an antisense mechanism resulting in loss of target RNA, we sought to directly compare siRNA-mediated with RNase H-mediated degradation of target RNA (12).
It has recently been reported that siRNA efficacy is highly dependent upon target position (36). Since RNase H-dependent oligonucleotides are also known to be dependent upon target position (34,41), siRNAs were designed to previously identified RNase H-dependent oligonucleotide binding sites to determine whether active RNase H-dependent oligonucleotide binding sites would be predictive of active siRNA sites. In three of four cases (ISIS 5132, ISIS 116847, and ISIS 16009), active siRNAs that targeted a site previously shown to be a good target site for RNase H-dependent oligonucleotides showed activity comparable with that of the RNase H-dependent oligonucleotide. One hypothesis for the lack of activity observed against the ISIS 2302 target, CD54, was that the siRNA mechanism is not amenable to silencing of TNF-␣-induced genes. This turned out not to be the case, since screening 40 siRNA molecules targeting different regions of the CD54 mRNA identified several active siRNAs.
Analysis of oligonucleotide screens against both CD54 and PTEN confirmed that target position is an important factor in determining siRNA activity. Our data suggest that there is an imperfect correlation between RNase H and siRNA oligonucleotide activity when they are designed to bind different regions of the target RNA. In general, sites on the target RNA that were not active with RNase H-dependent oligonucleotides were similarly not good sites for siRNA. Conversely, a significant degree of correlation between active RNase H oligonucleotides and siRNA was found, suggesting that if a site is available for hybridization to an RNase H oligonucleotide, then it is also available for hybridization and cleavage by the siRNA complex. However, some exceptions were noted, with sites identified that apparently were poor RNase H-dependent oligonucleotide targets but effective siRNA targets and vice versa. This dichotomy could be due to additional factors other than RNA accessibility, such as sequence preferences for the respective nucleases. Differences in activity between siRNAs and RNase Hdependent oligonucleotides may also result from structural differences between pre-mRNA and mRNA, which appear to be the targets for RNase H and siRNA oligonucleotides, respectively. Our data suggest that the secondary structure of the target RNA is an important determinant of activity for both siRNA and RNase H antisense oligonucleotides, and it can be assumed that the structure of a pre-mRNA containing intron sequences will be different from the structure of mature mRNA.
To determine whether siRNA molecules were more potent or effective inhibitors of gene expression in human cells, we compared an optimized siRNA molecule to an optimized 2Ј-MOE chimeric antisense molecule targeting either PTEN or CD54. In both cases, the oligonucleotides working by either antisense mechanism exhibited similar potencies in T24 cells. Additionally, both types of oligonucleotides inhibited the respective target genes by more than 90%. Some investigators have reported greater siRNA efficacy in cultured cells (9,36). However, others have reported activity comparable with what we report in this paper (11,42,43). One possible explanation for this difference would be that the siRNA molecules used in our studies were not optimally designed. For example, it has been demonstrated that a 5Ј-phosphate group is required for optimal siRNA activity (44). Since the siRNA oligonucleotides used in these experiments were not synthesized with 5Ј-phosphates, it is possible that greater potency would have been observed had the siRNA oligonucleotides been phosphorylated. However, published experiments (45) have revealed that there are no differences in efficiencies of 5Ј-phosphorylated and nonphosphorylated siRNAs in mammalian cells, since siRNA duplexes with free 5Ј-hydroxyls and 2-nt 3Ј overhangs are readily phosphorylated in the cell. Our own results confirm these observations (i.e. we did not see an increase in activity with siRNA molecules containing a 5Ј-phosphate) (data not shown). A more plausible explanation for the decreased siRNA potency in our study compared with others is the method chosen to quantify target reduction. Most other published studies have measured siRNA efficacy at the protein level. We chose instead to assay target reduction at the mRNA level using quantitative RT-PCR. Comparison of target protein and mRNA reduction demonstrates that several oligonucleotides appear to produce a more robust reduction of protein compared with the corresponding mRNA (Fig. 3, A and B). Additionally, the extreme sensitivity of quantitative RT-PCR compared with other assay methods may overrepresent RNA reduced to very low levels.
Both siRNA and the RNase H-dependent oligonucleotides gave similar duration of action in cultured cells, each showing a gradual recovery of mRNA expression over 4 -6 days. These results are in agreement with those reported previously for siRNAs (36) and second generation RNase H oligonucleotides (22). First generation phosphorothioate-modified oligodeoxynucleotides exhibit a duration of action and tissue half-life ranging from 24 to 48 h (46,47). In contrast, the second generation 2Ј-MOE-modified oligonucleotides used for the these studies exhibit a significant increase in nuclease resistance, resulting in a prolonged duration of action and tissue half-life from 5 to 10 days (22,23). If one takes into account the biostability of phosphorothioate-modified 2Ј-MOE antisense (23,48) and predicted stability of unmodified double-stranded RNA oligonucleotides, this result is somewhat surprising and suggests that the siRNA molecules may be protected from nucleases in cells. From this limited comparison, the onset of the RNase H-dependent activity appears to be slightly earlier than that of the siRNA. This may be a result of differences arising from the RNase H oligonucleotide acting in the nucleus on the pre-mRNA while siRNA acts cytoplasmically on the mature mRNA. Our data as well as other published data (49,50) suggest that the siRNA mechanism of action is restricted to the cytoplasm. In contrast, our results as well as previous publications (51,52) suggest that RNase H oligonucleotides are capable of binding to pre-mRNA in the cell nucleus. There may be specific applications in which it may be desired to utilize an RNase H oligonucleotides to inhibit all RNA variants derived from a single transcript or alternatively to selectively discriminate alternative spliced transcripts using siRNA in the cytoplasm.
The fidelity for perfect base pair matches for both types of oligonucleotides was investigated by designing oligonucleotides with internal or external 2-base mismatches. Activity was completely lost when 2-base mismatches were made in the central domain of either the RNase H-dependent oligonucleotide or siRNA. When mismatches were placed near the ends of the sequence, activity was reduced, but not completely. The loss of activity was greater for the RNase H-dependent oligonucleotide than the siRNA but not significantly so. Therefore, the two types of antisense oligonucleotides exhibit similar sequence selectivity.
In conclusion, we have compared RNase H-dependent antisense oligonucleotides with siRNA molecules targeting several human genes in cell-based assays. These studies have demonstrated that optimized siRNA and RNase H-dependent oligonucleotides behave similarly in terms of potency, maximal effects, specificity, and duration of action and efficiency. It remains to be determined whether siRNA molecules work broadly for in vivo applications. In a preliminary report of a siRNA molecule delivered to mice, the authors administered the oligonucleotide by rapid tail vein injection of a large volume of fluid (high pressure delivery) (53). It is not clear whether administration of siRNA molecules by more clinically acceptable practices will result in effective delivery to target tissues. In contrast, delivery of RNase H oligonucleotides to a variety of target tissues by a parenteral and nonparenteral routes of administration with subsequent inhibition of gene expression has been well documented in rodents, non-human primates, and humans (17, 23, 30, 32, 48, 54 -57). Both strategies, however, appear to be equally valid approaches for cell-based analysis of gene function in vitro.