siRNA Design: It's All in the Algorithm

Chris Echeverri, Christoph Sachse, Andrew Walsh, Anne Grabner Cenix BioScience GmbH
David Brown
Ambion, Inc.

RNA interference (RNAi) has become a powerful tool for understanding gene function. It is a cellular process wherein short double-stranded RNAs called short interfering RNAs (siRNAs) direct the degradation of transcripts containing sequence complementary to at least one of the siRNA strands [1,2]. siRNAs trigger RNA degradation through a protein complex referred to as the RNA Induced Silencing Complex (RISC) [3]. Most evidence indicates that the RISC contains only one of the two siRNA strands [4], suggesting that there is a step prior to, or during, the incorporation of the siRNA into the RISC that eliminates one of the siRNA strands [5]. Either siRNA strand can be taken up by the RISC [5,6], but the RISC can only direct degradation of cellular RNAs that are complementary to the bound siRNA. Recent work suggests that strand selection can be affected by the nucleotide composition of the siRNA [7,8]. It is therefore possible to select siRNA target sites that favor incorporation of the antisense siRNA strand into the RISC to increase the percentage of RISCs containing the correct targeting siRNA strand. This ultimately results in improved efficacy and specificity of the siRNA.

Maximize Success Rate of siRNAs
Publicly available siRNA design programs have so far shown success rates of 50-60% in generating siRNAs that can yield over 70% silencing of target mRNA levels in HeLa cells after 48 hr. Based on this benchmark, Cenix conducted a first test of its algorithm's success rate by measuring silencing efficacy under the same conditions, namely using quantitative RT-PCR to measure target mRNA levels in HeLa cells 48 hr after transfection, for a set of 79 algorithm-designed siRNAs, each of which targeted a different kinase transcript. The analysis revealed that 74 of the 79 siRNAs, or 94% of the tested siRNAs, gave higher than 70% silencing (see Figure 1).

Figure 1. Silencing Efficacy of 79 siRNAs Designed Using Cenix Algorithm to Target 79 Endogenously-expressed Human Kinases. Target mRNA levels were measured by qRT-PCR and normalized against 18S rRNA from samples harvested 48 hr after siRNA transfection into HeLa cells.

In subsequent analyses of sets of multiple siRNAs targeting the same transcripts, Cenix noted that certain genes, and in some cases certain transcripts, appear significantly more refractory to siRNA-based silencing than the majority (data not shown). Since it was not known what percent of genes would fall into this category, it became clear that any extrapolations of silencing success rates from tens or even hundreds of siRNAs would be of quite limited statistical value. Thus, expecting that the eventual "true" success rate of its algorithm would likely be lower than the 94% initially observed with the 79 kinases, Cenix extended the same performance analysis to a much larger set of siRNAs to create a more statistically relevant data set. Figure 2 shows the distribution of the siRNA efficacy for over 1,100 siRNAs targeting nearly 400 endogenously-expressed human transcripts.

The comprehensive survey indicated that when three Cenix designed siRNAs per gene were tested, one or more of the siRNAs achieved >70% silencing for over 93% of tested genes, >80% silencing for nearly 80% of tested genes, and >90% silencing for approximately half of tested genes. On a per siRNA basis, approximately 80% of the individual siRNAs showed >70% silencing of their target. The performance data clearly confirm the success of the Cenix siRNA selection process.

Maximizing Silencing Efficacy
In addition to increasing the likelihood that an siRNA will be active, the Cenix algorithm results in siRNAs that are remarkably potent (Figure 2). Using Cenix-designed siRNAs in optimized transfection systems, Ambion researchers routinely observe greater than 90% reduction in target mRNA levels with nM or even pM amounts of siRNA. Figure 3 shows an experiment in which siRNAs targeting three different genes were transfected at different concentrations. With one exception, each of the siRNAs performed essentially the same when transfected at either 200 pM or 10 nM. The observation that GAPDH siRNA was less effective at 200 pM than at 1 nM is probably because GAPDH mRNA is significantly more abundant in cells than the other two mRNA targets tested. It is probable that the concentration of GAPDH transcripts in the cell exceed the concentration of RISC-associated GAPDH siRNA generated by transfecting only 200 pM GAPDH siRNA.

Figure 2. Distribution of Gene Silencing Measured for 1,106 siRNAs Targeting 379 Endogenously-expressed Human Genes. Target mRNA levels were measured by qRT-PCR, normalized against 18S rRNA from samples harvested 48hr after siRNA transfection into HeLa cells. Percent of genes exhibiting siRNA-induced silencing above the noted thresholds (F70=70%, F80=80%, etc.) are shown.

Figure 3. Efficacy of Cenix-designed siRNAs. siRNAs targeting three different genes were transfected into HeLa cells. 48 hours post-transfection, RNAs from the cells were recovered and analyzed for target mRNA by real-time PCR, using 18S rRNA as the loading control. The reduction in target mRNA is calculated by comparing the PCR results for each test siRNA versus a negative control siRNA. Even extremely low concentrations of Cenix pre-designed siRNAs elicit a strong RNAi response.

There are several advantages to using low concentrations of highly active siRNAs; the most important is minimization of off-target effects, as these are known to increase with increasing concentrations of input siRNA [9, 10]. Off-target effects from high concentrations of siRNAs in the cells are likely to result because more of the sense strand becomes bound to the RISC and degrades transcripts with sequence complementary to the sense strand siRNA, or dsRNA binding proteins that stimulate antiviral response pathways are more likely to be activated and induce the expression of antiviral response genes. Another important advantage to using less siRNA for silencing experiments is that transfecting lower concentrations of siRNAs allows multiple siRNAs to be transfected at the same time. Ambion researchers have simultaneously reduced the expression of as many as five genes using highly active siRNAs designed with the Cenix algorithm (Figure 4).

Figure 4. Multigene Knockdown with siRNAs. Equal amounts of five different siRNAs were mixed and the indicated concentrations of siRNA were transfected into HeLa cells. 48 hours post-transfection, RNA from the cells was recovered and analyzed for target mRNA reduction by real-time PCR, using 18S rRNA as the loading control. The reduction in target mRNA is calculated by comparing the PCR results for each test siRNA versus a negative control siRNA at the same total siRNA concentration.

What Makes this Algorithm So Successful?
One theory broadly circulated early on in the field was that low overall G/C content in siRNAs tended to correlate with increased silencing efficacy. While it is true that effective siRNAs tend to have G/C contents less than 50%, overall G/C content is not very predictive of siRNA efficacy. Instead, Cenix recognized the strong impact of differential end-stabilities, as conferred by the choice of residues at or near the ends of the target sequence, on silencing efficacy. It emerged that siRNAs containing higher G/C contents at certain residues near the 3' end of the antisense strand, and lower G/C content at certain residues near the 5' end of the antisense strand offered significantly higher success rates. This principle was therefore implemented as one of the key algorithm determinants for maximizing silencing efficacy.

Since then, a recent article published in the journal Cell has suggested a biological reason for the correlation between the nucleotide end-composition of the target site and siRNA efficacy. Schwarz et al. [7] used an in vitro assay system to demonstrate that the sense and antisense strands of an siRNA are not equally likely to be bound by the RNA Induced Silencing Complex (RISC). Each RISC in a cell uses only one strand of an siRNA as a guide for RNAi [4], thus the strand that is bound by RISC dictates what mRNA sequences are targeted for degradation. Schwarz et al. noted that the strand of the siRNA whose 5' end had a lower G/C content was preferentially loaded. In fact, siRNAs with high duplex stability at one end and low duplex stability at the other exhibited such significant strand bias that one strand could be incorporated into RISC to the exclusion of the other strand. The authors hypothesized that an RNA helicase responsible for unwinding siRNAs selects a strand for incorporation into the RISC based on the ease with which it can unwind the first 4-5 nucleotides of the duplex (Figure 5). These findings help explain the sequence bias between effective and ineffective siRNAs that was observed by Cenix. An siRNA target site with high G/C content at positions 1-4 and low G/C at positions 16-19 results in an siRNA whose antisense strand (the strand complementary to the desired mRNA target) has a low G/C at the 5' end and a high G/C at the 3' end. According to Schwarz et al., the antisense strand would be preferentially incorporated into RISC and target the proper mRNA for degradation.

Preferential uptake of the siRNA antisense strand by RISC has two important consequences: (1) The siRNA tends to have higher efficacy since the correct strand is efficiently taken up by the enzyme complex responsible for mRNA degradation and (2) the siRNA has higher specificity since the sense strand is not taken up by RISC and thus cannot guide the degradation of mRNAs with sequence elements that are at least partially complementary to the siRNA target site [9].

Going Beyond Terminal T^_m
In addition to G/C content at the termini of the siRNAs, traits that were found to be influential in defining optimal siRNA sequences include the T^_m of specific internal domains of the siRNA, siRNA length, position of the target sequence within the CDS (coding region) and nucleotide content of the 3' overhangs. These findings are incorporated into the Cenix algorithm used to design Ambion's Silencer™ Pre-designed siRNAs and Silencer™Validated siRNAs as well as the Silencer™ siRNA Libraries. The design process also includes a specificity check whereby both siRNA strands are subjected to customized homology searches to minimize risks of generating off-target effects, and putative siRNA target sequences are screened against the most updated single-nucleotide polymorphism (SNP) databases to avoid variability in siRNA efficacy. Target mRNA sequences known to mediate regulatory processes through binding to protein factors are similarly avoided, as are CpG motifs.

Concentrate on Your Experiments, Not on your siRNA Design
Experiments at Cenix BioScience, Ambion, and elsewhere [7,8] have shown that design rules exist for siRNAs, and that they can be exploited to both improve the chances that an siRNA will be active and enhance the efficacy of the siRNAs. Ambion's Silencer Pre-designed Validated and Library siRNA products eliminate the need to screen siRNAs for functionality, so that researchers can concentrate on their experiments and avoid wasting time on reagent design.

back to top

Ordering Information for Ambion Products: