Non-specific Gene Silencing by Long dsRNAs
While the natural presence of RNAi had been observed in a variety of organisms (plants, protozoa, insects, and nematodes), evidence for the existence of RNAi in mammalian cells took longer to establish. Transfection of long dsRNA molecules (>30 nt) into most mammalian cells causes nonspecific suppression of gene expression, as opposed to the gene-specific suppression seen in other organisms. This suppression has been attributed to an antiviral response, which takes place through one of two pathways.

In one pathway, long dsRNAs activate a protein kinase, PKR. Activated PKR, in turn phoshorylates and inactivates the translation initiation factor, eIF2a, leading to repression of translation. (39) In the other pathway, long dsRNAs activate RNase L, which leads to nonspecific RNA degradation (40).

A number of groups have shown that the dsRNA-induced antiviral response is absent from mouse embryonic stem (ES) cells and at least one cell line of embryonic origin. (41, 42) It is therefore possible to use long dsRNAs to silence specific genes in these specific mammalian cells. However, the antiviral response precludes the use of long dsRNAs to induce RNAi in most other mammalian cell types.

siRNAs Bypass the Antiviral Response
Interestingly, dsRNAs less than 30 nt in length do not activate the PKR kinase pathway. This observation, as well as knowledge that long dsRNAs are cleaved to form siRNAs in worms and flies and that siRNAs can induce RNAi in Drosophila embryo lysates, prompted researchers to test whether introduction of siRNAs could induce gene-specific silencing in mammalian cells (43). Indeed, siRNAs introduced by transient transfection were found to effectively induce RNAi in mammalian cultured cells in a sequence-specific manner. The effectiveness of siRNAs varies — the most potent siRNAs result in >90% reduction in target RNA and protein levels (44-46). The most effective siRNAs turn out to be 21 nt dsRNAs with 2 nt 3' overhangs. Sequence specificity of siRNA is very stringent, as single base pair mismatches between the siRNA and its target mRNA dramatically reduce silencing (44, 47). Unfortunately, not all siRNAs with these characteristics are effective. The reasons for this are unclear but may be a result of positional effects (46, 48, 49). For current recommendations on designing siRNAs, see "siRNA Design".