The first in a two-part series on RNAi
Carolyn Riley Chapman, PhD
There is no single, easy solution for siRNA design and delivery, but tools are being developed which could be applied on a case-by-case basis.
The use of RNAi as a research tool for target identification and validation, though relatively new, has now become common practice at pharmaceutical and biotechnology companies. Its early promise will not be fully realized, however, until remaining technical hurdles are overcome, particularly in the area of in vivo delivery methods.
RNAi has its roots in genetics (the phenomenon of RNA-induced gene silencing was first observed in plants and worms), and drug discoverers still rely on model organisms for target identification. For example, researchers at Bristol Myers Squibb Co. (BMS), Princeton, N.J., have collaborated with Exelixis, South San Francisco, Calif., to conduct RNAi-based genetic screens in Caenorhabditis elegans to identify oncology drug targets.
As techniques and tools have evolved, RNAi screening for target identification and validation is increasingly being carried out in mammalian cells. In the last few years, C. elegans and Drosophila have been phased out, says Christophe Echeverri, PhD, CEO and chief scientific officer of Cenix BioScience, Dresden, Germany. Cenix now routinely conducts its siRNA screens in human and rodent cell-culture systems, he says. To make the jump from model organisms to mammalian cells, the first challenge was creating potent RNAi libraries. Cenix also spent a great deal of time optimizing transfection protocols for a wide range of cell lines, not always an easy task.
The type of screen being performed has also shifted over time. While a lot of early experiments involved identifying siRNAs that caused “life or death of cells,” says Eric Lader, PhD, associate director for R&D at Qiagen, Inc., Germantown, Md., many scientists have moved on to more sophisticated and higher content assays such as high-throughput fluorescence microscopy. One key reason: screens that are not specific enough generate too much noise. “For pharma and biotech, there’s no way they can follow up on a thousand hits in a genome-wide screen—it’s just too expensive,” says Lader, so they are spending more time upfront developing assays and performing primary screens.
siRNA design
Discovering Dicer’s Crystal Structure Although exploitation of cellular RNAi pathways is now commonplace for target identification and validation in drug discovery, researchers are still unraveling the RNAi pathways themselves. One recent advance was the determination of the X-ray crystal structure of the Dicer enzyme from Giardia intestinalis by a group of researchers at the University of California at Berkeley/Howard Hughes Medical Institute, led by Jennifer Doudna, PhD. Dicer cleaves double-stranded RNA (dsRNA) into smaller siRNA and miRNA segments that cause degradation or translational repression of target mRNAs. The crystal structure clearly showed that the dimensions of Dicer match the length of an siRNA, says Ian MacRae, PhD, postdoctoral fellow and lead author of the study, which was published in Science. Dicer has a PAZ domain that recognizes the ends of dsRNA. Sixty-five angstroms away (the length of 25 nucleotides), it has two RNase III domains that cut the dsRNA. Between the two regions is a flat-surfaced, positively-charged region that probably facilitates binding to negatively charged dsRNA. Dicer is “like a ruler-one end is a clamp and the other end is a pair of scissors,” MacRae says. After being cleaved, siRNA and miRNAs are incorporated into RISC. In 2004, Argonaute was identified as RISC’s enzymatic core when its crystal structure revealed similarity to RNase H. |
Understanding how RNAi pathways work in cells has greatly informed RNAi experimentation. Perhaps this is most evident in the design of siRNAs. Most people choose synthetic siRNAs for their experiments because it is the easiest, most cost effective route, says Peter Welch, PhD, director of gene regulation at Invitrogen Corp., Carlsbad, Calif.
To come up with potent siRNA reagents, computer algorithms select for sites within genes that look favorable for silencing, says Mark Behlke, MD, PhD, vice president of molecular genetics & biophysics at Integrated DNA Technologies (IDT), Coralville, Iowa. Through the work of many different labs, certain features of small RNAs have been identified and validated as contributing to their silencing activity. The algorithms use those parameters to estimate an siRNA’s ability to silence its intended target, as well as its potential for off-target effects.
Andrew Peek, PhD, director of bioinformatics at IDT, says the preeminent feature is position-specific nucleotide composition of the small RNAs. “It turns out that most of the specificity and binding energy of a small RNA comes from just six or seven nucleotides of the 21, a region called the seed sequence,” says Phillip Zamore, PhD, professor at the University of Massachusetts Medical School, North Worcester, Mass. “That totally changes expectations for specificity and changes how you would go about designing RNAs and analyzing them,” Zamore adds, emphasizing that small RNAs do not function like antisense oligonucleotides. “If they had worked like that, this would all be so much easier.”
Another major factor the algorithms weigh is the thermodynamics between the guide and passenger strand, says Peek. Groups led by Zamore and Anastasia Khvorova, PhD, at Dharmacon, Inc., Lafayette, Colo., discovered that double stranded RNA is loaded asymmetrically into RISC, the RNA-induced silencing complex. “One end is simply less tightly paired than the other, and that end uniquely identifies the strand that will be loaded into RISC, and the other strand gets destroyed,” explains Zamore. “From the standpoint of target validation and therapeutics, it changed the way we think about designing a small RNA,” Zamore says. It is crucial to design siRNA duplexes so the appropriate strand will enter the silencing pathway.
Lader says that state-of-the-art algorithms now produce siRNA oligos with an 80% success rate, with success typically defined as 70% or greater attenuation of the mRNA. That means that 80% of oligos randomly generated against 100 different genes will “work.” Extrapolating from that number implies that there is a 99% chance that at least one of three different siRNA oligos will work against any particular gene. “The problem is that you can’t extrapolate that number . . . some genes are easier to knock down than others,” says Lader.
One of the most important aspects of assay development is setting the threshold for defining positives and negatives in a screen. “If you set the baseline too high, you miss a lot of true hits. If you set the baseline too low, you catch all your true hits but you have a lot of noise,” says Lader. But even after a reasonable baseline is achieved, screeners still worry about false negatives and false positives.
“One way to get a false negative is an siRNA that doesn’t work,” Lader adds. In fact, Echeverri believes that the most important role of the silencing reagent is that “it silences very, very well so that you minimize your chance that you may have just not gotten to the threshold that can give you a loss-of-function phenotype.”
Hitting the right targets
Optimizing computer algorithms and wet lab validation of RNAi knockdown are important aspects of making sure the reagents are knocking down their intended targets. At Cenix, the first pass of screening is usually performed with siRNAs at high concentrations to reduce the chance of false negatives, Echeverri says, with secondary screening at lower siRNA concentrations to confirm results.
“False positives are always a problem and always occur,” says Haney. To ensure that an observed phenotype is truly caused by down-regulation of the intended target, Wyeth looks at titration series and time courses to see if the observed phenotypes correlate with the profile of RNA knockdown. Cristina Rondinone, PhD, director of research for metabolic diseases, at Hoffmann-La Roche Inc., Nutley, N.J., also recommends using the lowest concentration of siRNAs required for a particular assay to avoid spurious results. “Sometimes when you go higher you can have some off-target effects.”
Issues of redundancy, protein thresholds, and mRNA and protein turnover also complicate data interpretation. There may be compensatory mechanisms that circumvent the loss of one protein, for example. Alternatively, there is variation in the threshold of knockdown needed to get a phenotype. Knocking down some mRNAs by just 20% or 30% leads to profound phenotypes, while other proteins may function sufficiently with just 10% of their normal mRNA levels.
To be sure knockdown of a particular target is causing the observed phenotype, researchers also utilize several independent siRNAs. “Minimally, we generate four siRNAs,” says Lata Jayaraman, PhD, senior research investigator in oncology drug discovery at BMS, and “if a target starts to look interesting, we go ahead and make more siRNAs to increase our confidence level.” Drawing conclusions from the data, however, requires careful judgment. For example, if four of eight siRNAs cause strong protein knockdown, but do not share the same phenotype, it can be puzzling, Jayaraman says. But Echeverri maintains that even if only two different siRNAs to a particular gene cause the same specific phenotype, it’s difficult to dismiss the observations as artifacts.
Rescue experiments are also important for determining specificity of effect, says Echeverri. One way rescue experiments are performed is by supplying another version of the gene that shouldn’t be silenced by a given siRNA. For example, one could use the mouse orthologue in human cells and pick an siRNA sequence that targets the endogenous human gene specifically. Complementary DNA constructs are the most obvious way to perform these types of experiments, Echeverri says, but the better approach is to use bacterial artificial chromosome constructs, where the rescue occurs with more physiological levels of expression.
A number of experts also point to a trend towards primary screening with one siRNA per well, since pooling siRNAs can generate both false negatives and false positives.
Vector-based RNAi
There are two main reasons scientists move from siRNAs to vector-based systems for their RNAi experiments: delivery and sustained silencing. “shRNA and miRNA are mostly used for focusing on one or a few genes, or developing specific cell line models that would require stable expression of the knockdown, or in primary cells or other cell types that cannot be readily transfected by siRNA,” explains Haney.
shRNAs are short hairpin RNAs that contain sense and antisense sequences connected by a stem loop. A polIII promoter, usually U6 or H1, drives transcription of shRNA from a vector that may or may not become integrated into the genome. shRNA has been the primary choice for vector-based RNAi knockdowns to date, but experts in the field relay excitement about newer miRNA vector systems. miRNA refers to the naturally-occurring, non-coding RNAs that are believed to regulate gene expression, usually through blockage of protein translation from one or more target mRNAs. Scientists are now making vector-based RNAi systems that mimic endogenous miRNA structures.
“You can now take gene-specific miRNAs in an authentic miRNA backbone and target your gene of choice,” explains Haney, who is an adopter of miRNA technology. “You can actually also insert shRNAs if that’s what you want to do.” For example, Invitrogen started with mir155, a naturally occurring microRNA, and changed the 21-base pair targeting sequence to recognize other targets of choice. Like siRNAs, shRNAs lead to RNA degradation, so their effects can be measured in TaqMan assays that quantify mRNA. But since miRNAs regulate mRNA expression by blocking protein translation initiation, their effects can only be seen at the protein level.
According to Haney, although miRNAs are similar to shRNAs, one key difference is that they run off polII promoters, which allow researchers to take advantage of “very robust, tissue specific and regulatory modifications.” There are some well-reported examples of regulated knockdowns of shRNAs from polIII promoters, Haney says, but in general, he believes it’s more hit or miss. According to Welch, the success rate of achieving RNA knockdown is also higher with miRNAs than shRNAs, and miRNAs can also be used at lower concentrations, possibly because the cellular machinery more easily recognizes the miRNA structures.
In vivo delivery
Once an siRNA and target gene of interest are identified, many researchers want to perform validation experiments in vivo. But many scientists would like to go directly in vivo using the same siRNAs that were used for in vitro experiments. In doing this, researchers have encountered some hurdles. The first was stability, but that problem has been largely solved. Several companies, including Alnylam Pharmaceuticals Inc., Cambridge, Mass., have developed modifications to siRNAs that make them more stable in plasma or serum. The first, the incorporation of a single phosphorothioate linkage at the 3′ termini on each strand, protects the siRNA from exonuclease degradation; 2′-O-methyl and/or 2′-fluoro modifications can also be added to protect sites vulnerable to endonuclease cleavage, says Nagesh Mahanthappa, PhD, senior director of business development and strategy at Alnylam.
A more challenging obstacle to in vivo experimentation with siRNAs has been delivery. The penultimate goal would be to achieve a universal systemic delivery strategy for siRNAs. On this front, Mahanthappa says Alnylam has published that covalent conjugation of cholesterol to an siRNA molecule increases its circulating half-life and promotes uptake into a variety of cell types. Alnylam is also actively investigating other types of formulation approaches, such as mixing siRNAs with lipid components, to promote uptake by different cells.
That said, there doesn’t seem to be a single, easy solution for systemic siRNA delivery, he says. “You’re not going to see a one-size-fits-all strategy, even from a target validation perspective,” says Mahanthappa. “I think there will be favored types of chemical modifications or formulations that will allow one to deliver siRNAs to particular cell types. As those are developed, it’s going to give rise to a toolkit or a palate of methodologies which can then be applied on a case-by-case basis.”
About the Author
Chapman is a freelance writer based in Harrison, N.Y.
This article was published in Drug Discovery & Development magazine: Vol. 9, No. 7, July, 2006, pp. 37-42.
Filed Under: Drug Discovery