YouTube Facebook LinkedIn Google+ Twitter Xingrss  

By Dmitry A. Samarsky and Peter J. Welch

December 15, 2004 | Traditionally, development of novel technologies and research tools follow the basic scientific discoveries. The RNAi story is a bit different. As a new tool for gene knockdown in mammalian cells, RNAi was introduced to the public in mid-2001. In just three years it became one of the most widely used technologies in both academic and industrial research environments. This unprecedented implementation speed can be explained by an urgent need in the scientific community to decipher sequence data generated by human genome projects during the previous years.

Mammalian scientists were finally given an easy and robust way of knocking down a gene without having to use time-consuming and less efficient methods such as homologous recombination, antisense, ribozymes or antibody microinjection. In its turn, such a rush created a unique situation: RNAi applications started to precede understanding of the underlying mechanisms. Indeed, researchers started using RNAi compounds to knockdown genes, with high success rates, without even knowing exactly how theses compounds were causing the destruction of target mRNAs. Following the demand of the scientific community to generate better and more affordable RNAi products, we and others started to dig deeper into the principles of RNAi.

One inherent advantage of RNAi is that to inactivate mRNA targets it uses the cell's own knockdown machinery. Previous RNA knockdown technologies, such as antisense and ribozymes, generated a great deal of excitement early in their discovery. However, once reduced to general practice, these methods were found to be less than ideal. Identification of good antisense or ribozyme targets was found to be heavily dependent on the quality of the design algorithms, often exquisitely sensitive to secondary structure in the target mRNA, and often requiring the researcher to screen a large number of target sites before finding an effective one.

In addition, many successful knockdown experiments required relatively high doses of antisense or ribozyme molecules to show reasonable effectiveness. RNAi is proving to be considerably more robust and effective, requiring very few targets to be screened before finding ones that work at low concentrations.

One of the key factors contributing to the effectiveness of RNAi in mammalian cells is the development of good design rules. Several seminal publications described an interesting correlation: siRNAs with weak (A-U) base-pairing at the 5' end and strong (G-C) base-pairing at the 3' end of the duplex (corresponding to the 5' and 3' ends of the antisense strand, respectively) performed better (Khvorova et al. 2003; Schwarz et al. 2003) and internal data generated in our labs supported these findings. It was found that the naturally occurring RNAi cellular machinery, referred to as RISC (RNAi-induced silencing complex), was selective in choosing the ends of siRNAs, preferring the ends with weak Watson-Crick base-pairing. By properly exploiting this property, it is now possible to design synthetic RNAi compounds with success rates greater than 90 percent.

Unlike antisense and ribozymes, RNAi effectiveness appears to be fairly insensitive to the structure and context of the target mRNA. Early algorithms included mRNA folding calculations, with the goal of targeting regions of open secondary structure. Remarkably, we and others found that this step can be eliminated completely without affecting the efficiency of designed siRNAs. It has been postulated that the RISC-associated helicases may help unwind target mRNA secondary structure making them accessible to the incoming RNAi. The benefit of eliminating the folding portion of the algorithm is that it increases the number of potentially good target sites

RNAi also seems to be unaffected by the context of the target site. For example, we have developed an RNAi screening vector in which the target gene is placed downstream of a reporter gene (e.g., lacZ) to create a fusion mRNA. The relative effectiveness of a large number of siRNA target sites can be rapidly and quantitatively determined in any easy-to-transfect cell line. We and others have found a very strong correlation between data generated with this screening vector and data generated against the naturally-occurring endogenous target mRNA. In other words, the context of the target site does not seem to affect RNAi activity and, if an RNAi compound works against one transcript, it will work against any other transcript containing the same target sequence.

We found yet another striking difference between RNAi and antisense compounds when we tried to establish the procedure for visualization of RNAi compounds after transfection (with lipids or using electroporation) inside the mammalian cells. While fluorescently labeled DNA-based antisense oligonucleotides are readily seen under the microscope, RNAi compounds either generate confusing "punctate" patterns or are undetectable altogether, depending on the type of labeling and delivery. Remarkably, these "undetectable" RNAi compounds manage to generate robust target cleavage, sustainable for at least several days. Detailed analysis of this phenomenon lead to a conclusion that upon transfection most of the cytoplasmic siRNA fraction gets quickly degraded, and the RISC retains only relatively small portion of antisense strands. Such "charged" RISC is thought to remain highly active in multiple rounds of cleavage.

Stealth Mode 
Analysis of another finding helped us to further refine the picture and to improve siRNA chemistry. In our effort to develop the next-generation RNAi compounds, we applied chemical modifications to improve the specificity and stability of the original siRNAs. The resulting Stealth RNAi compounds exhibit strand specificity, in which only the antisense strand can be utilized for RNAi, thus reducing the chance of off-target effects. In addition, the modifications on the Stealth RNAi molecules avoid the interferon/stress responses that can confound the interpretations of experiments using standard siRNAs (Kim et al. 2004; Sledz et al. 2004).

Finally, these chemical modifications resulted in RNAi molecules that are highly resistant to nucleases in blood serum. While the increased stability of Stealth in serum makes it particularly attractive for in vivo applications, it does not significantly prolonged target knockdown within the cell. Instead, we have found that highly effective Stealth molecules remain active considerably longer in the cell (up to seven days) as compared to less effective duplexes that can disappear within three days. Together these data suggest that RNAi effectiveness is linked to RISC longevity rather than simply RISC availability.

A Simple Model 
The function of RISC is a critical component of the mechanism of RNAi. It defines which strand of RNAi compound to incorporate, how efficiently it is taken up, and how long it is active against target mRNAs. Taken together these observations argue for the following simple model. Upon delivery inside the cell, a small portion of the RNAi antisense strand gets selectively incorporated into the RISC complex, which becomes extremely potent (most likely working in multiple cleavage rounds). This indicates that the difference between RNAi and traditional antisense is far from being just semantic, as it might have looked initially; the difference is fundamental. With antisense, the thermodynamics of the complementary bond formation define the efficacy of the process. With RNAi, the RISC incorporation and effectiveness of the antisense strand is what differentiates a good siRNA from a poor one.

The question to be answered now is how RISC, with incorporated appropriate antisense strand, manages to be so highly effective, targeting nearly every target mRNA in the cell. An attractive possible explanation is as follows. The free RISC is loose and is fishing in cytoplasm for RNAi compounds to incorporate one of the strands. Once "charged" with the RNAi strand, the RISC occupies a position through which every mRNA transcript has to pass (this could be at the nuclear pore complex, at the ribosomes, or elsewhere in the cell), something like a tunnel with the RNAi antisense strand exposed for interaction with the target. This way every transcript in a linearized form passes through the tunnel and has a chance for interaction with the antisense strand, which guarantees high efficiency and accuracy of the process.

Dmitry A. Samarsky is responsible for business and technology development at Invitrogen. Peter J. Welch is associate director of R&D at Invitrogen. 

Additional Reading 

Khvorova A, Reynolds A and Jayasena SD (2003). Functional siRNAs and miRNAs exhibit strand bias. Cell 115: 209-216.

Kim DH, Longo M, Han Y, Lundberg P, Cantin E and Rossi JJ (2004). Interferon induction by siRNAs and ssRNAs synthesized by phage polymerase. Nat Biotechnol 22: 321-325.

Schwarz DS, Hutvagner G, Du T, Xu Z, Aronin N and Zamore PD (2003). Asymmetry in the assembly of the RNAi enzyme complex. Cell 115: 199-208.

Sledz CA, Holko M, de Veer M, Silverman R and Williams BRG (2003). Activation of the interferon system by short-interfering RNAs. Nat Cell Biol 5: 834-839.

Back to Running Interference 

For reprints and/or copyright permission, please contact  Jay Mulhern, (781) 972-1359,