Selection of RNAi reagents for genomic screens: siRNA and shRNA

Thomas Murphy, PhD

When I first entered the field of functional genomics in 2003, the potential of RNA interference (RNAi) astonished me. As a scientist who was schooled in classical hypothesis-driven research, the concept of performing >10,000 parallel mammalian cell siRNA transfections in a 384 well plate format seemed unattainable. However, as time proceeded, I gained a greater appreciation of the power of reverse-genetics in which a gene of interest is targeted for disruption and the impact is measured by a phenotypic assay. Until recently, loss-of-function analysis in mammalian cells was difficult and time-laborious, requiring the creation of knock-out cell lines or organisms, and thus greatly limited its experimental utility. RNAi, thru its ability to systematically silence any gene of interest, allowed the classic reverse-genetic approach to be applied to a mammalian system in a time and cost-effective manner. Once mated with high throughput technologies, RNAi provided an attractive strategy to functionally annotate the whole human genome.

The RNAi mechanism

As a field of study, RNAi is relatively young. The original report of the RNAi effect in C. elegans was published just 8 years ago by Andrew Fire and Craig Mello, both of whom were recently awarded the Nobel Prize in Medicine for their contributions. Fire and Mello observed that by delivering long double stranded RNA (dsRNA) into the worm, they could trigger potent gene knockdown [1]. Since that original report, the field of RNAi has progressed rapidly. We now know that those original long dsRNA are cleaved by the ribonuclease Dicer to yield a pool of short interfering RNAs (siRNAs). These siRNAs then associate with the RNAi Induced Silencing Complex (RISC) which unwinds the two strands and loads the guide strand into the active site. The activated complex then searches the transcriptome for sequences of complete homology and subsequently mediates the binding and cleavage of target mRNA. While siRNAs can be produced in vivo using dsRNA triggers, their tendency to induce the interferon (IFN) response (and thus cell death) precludes their use in mammalian cells. Tom Tuschl’s group then at the Max Planck Institute for Biophysical Chemistry was one of the first to recognize that transfection of short synthetic molecules could simultaneously bypass IFN induction and elicit potent gene silencing [2]. Expression of short hairpin RNAs (shRNA) by transient transfection of a plasmid or implementation of a viral delivery system can also be employed [3].  In these systems, Dicer (or a combination of both Drosha and Dicer depending on the hairpin design) cleaves the shRNA into a mature siRNA (Figure 1).

Figure 1: The RNAi mechanism

The application of RNAi to screens

Historically, new technologies often transition from single gene applications to large gene sets. For example, PCR was initially applied to small gene collections but was later adapted to genome-wide applications in conjunction with microarray technologies. Expansion of RNAi technology has followed this evolution with initial reports focusing on small scale, single gene applications. Within just two years, libraries targeting entire gene families were reported[4] and more recently an arrayed lentiviral shRNA library targeting 5000 genes was used to identify novel mitotic regulators [5]  

Implementation of a large-scale RNAi screen is an ambitious goal which can greatly benefit from meticulous experimental design. Careful attention to the type of silencing reagent employed, the cell type being used, the assay configuration, and the statistical methods used to assess data quality and hit validity, is critical. In this article, I intend to focus on the issue of reagent selection. As a field application scientist, I have the opportunity to interact with a large number of RNAi users and would like to share their observations concerning the choice of synthetic and vector-based RNAi in high throughput screening.  

Choosing the ideal reagent

The configuration of the silencing reagent can determine how a screen is designed and implemented. When talking with investigators who are new to screen design (and RNAi in general), I frequently find a fair amount of confusion and misinformation over the strengths and weaknesses of synthesized siRNAs and their virally-delivered counterparts, short hairpin RNAs (shRNAs). From the larger perspective, siRNAs, by virtue of the fact that 1) they are manufactured chemically, 2) can be chemically modified to enhance specificity, and 3) are provided in a ready-to-use format, are much easier to put into practice, particularly when scaling up to perform one or more genome-wide screens. In contrast, a substantial production cost, maintenance infrastructure, and custom automation is associated with high-throughput viral-based screening [5]. For whole genome coverage, greater than 100,000 bacterial stocks encoding the shRNA constructs require maintenance, curation, and expansion to yield purified plasmid DNA for viral particle production. Because these processes require greater biosafety measures than synthetic siRNAs, from a purely practical standpoint, siRNAs are more easily adapted to the high throughput, automated environment of a genomic screening.

Functionality

Considerable effort has been invested into the development of algorithms that enhance the functionality of synthetic siRNA molecules [6].  The resulting siRNA rational design algorithms select for desirable thermodynamic profiles and positional base pair preferences, and eliminate molecules that contain unfavorable secondary structures and GC contents. Today’s most advanced algorithms are robust and are able to select for siRNAs exhibiting mRNA knockdown efficiencies of 75% or greater (e.g., Dharmacon siGENOME siRNA).

In contrast, knockdown potency has been a major limitation for shRNA constructs. This stems from the complexity associated with Drosha and Dicer processing of shRNA molecules and the absence of sophisticated algorithms for unimolecular designs. As an example, a recent sub-genome scale screen utilizing lentiviral delivered shRNAs found that only 30% of shRNAs knocked down the intended target by 75% or greater [5]. Taking this into account, 5-10 viral constructs are required per gene target in order to yield 2-3 functional shRNAs (the minimum required to show functional redundancy).Thus, for even a small scale screen comprising 500 gene targets, this would require the coordinated production of 2500-5000 viruses with equivalent titers. Therefore, from the perspective of functionality, synthetic duplexes are more likely to yield robust and consistent knockdown than their expressed shRNA counterparts.

Specificity

As early as 2003, it was clear that siRNAs could trigger down regulation of unintended gene targets [7, 8] and that in some cases, these events could yield detectable cellular phenotypes [9]. These aberrant phenotypes are typically referred to as false positives and potentially confound the interpretation of gene functional analyses and hit identification [10]. The source of off-target effects remained largely unknown until a recent publication provided compelling evidence for a link between siRNA off-targeting and microRNA (miRNA) targeting. miRNAs are the endogenous substrates of the RNAi pathway and act through pairing of the guide strand seed region (nucleotides 2-7) and complementary sequences in the 3’ UTR of the targeted gene.  Oddly enough, siRNA off-targets exhibit the same biases, thus hinting of a similarity in the mechanism between the two types of gene targeting[11].

Intent on solving this problem, Devin Leake at Dharmacon recently identified a chemical modification pattern that greatly reduced siRNA off-target effects. Described in a 2006 publication with Rosetta Inpharmatics, the modifications simultaneously eliminate sense strand interactions with RISC and minimizes antisense strand annealing with targets having less than 100% complementarity [12]. Extensive experimentation has shown that these modifications provide an advanced level of specificity compared to unmodified single siRNA duplexes, particularly in the context of a pool of four individual duplexes. Moreover, the modification pattern identified by Leake has been combined with additional specificity-enhancing bioinformatic tools to generate the most recent generation of highly specific, highly functional silencing molecules, the ON-TARGETplus™ line of siRNA reagents. Unfortunately, the same modifications cannot be incorporated into shRNA generated from expressed sequences, thus limiting the set of tools that can be applied to prevent off-target effects generated by vector-based constructs. Moreover, Mark Kay’s group at Stanford has recently demonstrated that virally delivered shRNAs can interfere with the cell’s ability to process its own endogeneous miRNA precursors [13]. As these molecules play a role in regulating as much as a third of the genome, non-specific interference with the pathway could result in a wide array of false positive phenotypes. Thus, for reasons of specificity and interference with endogenous processes, synthetic molecules are more applicable to RNAi-based HTP screens than expressed hairpin counterparts.

Delivery and Silencing Duration

Without question, one major advantage of viral vectors is their ability to infect cells that are intractable to lipid-mediated siRNA transfection. Lenti- and retroviruses also have the added ability to confer stable knockdown through the integration of the shRNA expression construct into the host cell’s genome. These attributes can be very advantageous when you are dealing with proteins which have a very long half life or wish to investigate the impact of silencing over weeks to months rather than days. siRNAs generally exhibit silencing potential for 5-7 days, dependent on the rate of cell doubling and target protein turnover. Chemical modifications can also be applied to the siRNA during synthesis that enhances stability, especially for in vivo applications (for instance siSTABLE® modification). Nonetheless, for some assays, this length of silencing is not sufficient. In the context of high throughput screening, it is important to ask the question: Do I need long term silencing for my application? In those settings where the timeframe of the experiment will be completed over the course of a few days, then clearly the choice to use synthetic siRNAs is advised.

Summary

With Tuschl’s watershed discovery that RNAi mediated knockdown could be performed in mammalian cells, a new phase in genome wide screening was ushered in. Clearly, siRNAs and shRNAs have distinct advantages and disadvantages and represent complementary technologies. I’d suggest deciding a priori how best to utilize the relative strengths of each reagent. For instance, one might start by designing an RNAi genomic screen utilizing a chemically synthesized siRNA library and a transfectable cell type that most closely mirrors the biological model of interest. From the screen you will then be able to prioritize a shorter list of candidate genes to perform further mechanistic studies. If, in these follow up assays, transfection and/or duration of knockdown require a viral shRNA solution, then it will be much easier to utilize the technology at this stage in the screening workflow with a more manageable number of viral constructs. By adopting this strategy, your screen can benefit from the advances made in siRNA design (i.e., functionality, specificity, and feasibility) and the advantages of shRNA (delivery into hard to transfect cell types, silencing duration). In this way, the application of these technologies to reverse-genetic experiments promises to unlock many of the genome’s mysteries.  Happy gene hunting!

References:

1. Fire, A., et al., Potent and specific genetic interference by double-stranded RNA in Caenorhabditis elegans. Nature, 1998. 391(6669): p. 806-11.
2. Elbashir, S.M., et al., Duplexes of 21-nucleotide RNAs mediate RNA interference in cultured mammalian cells. Nature, 2001. 411(6836): p. 494-8.
3. Paddison, P.J., et al., Short hairpin RNAs (shRNAs) induce sequence-specific silencing in mammalian cells. Genes & Dev., 2002. 16(8): p. 948-1.
4. Brummelkamp, T.R., R. Bernards, and R. Agami, A system for stable expression of short interfering RNAs in mammalian cells. Science, 2002. 296(5567): p. 550-3.
5. Moffat, J., et al., A lentiviral RNAi library for human and mouse genes applied to an arrayed viral high-content screen. Cell, 2006. 124(6): p. 1283-98.
6. Elbashir, S.M., et al., Functional anatomy of siRNAs for mediating efficient RNAi in Drosophila melanogaster embryo lysate. EMBO Journal, 2001. 20(23): p. 6877-88.
7. Jackson, A.L., et al., Expression profiling reveals off-target gene regulation by RNAi. Nature Biotechnology, 2003. 21(6): p. 635-7.
8. Semizarov, D., et al., Specificity of short interfering RNA determined through gene expression signatures. Proceedings of the National Academy of Sciences U S A, 2003. 100(11): p. 6347-52.
9. Fedorov, Y., et al., Off-target effects by siRNA can induce toxic phenotype. RNA, 2006. 12(7): p. 1-9.
10. Lin, X., et al., siRNA-mediated off-target gene silencing triggered by a 7 nt complementation. Nucleic Acids Res, 2005. 33(14): p. 4527-35.
11. Birmingham, A., et al., 3' UTR seed matches, but not overall identity, are associated with RNAi off-targets. Nat Methods, 2006. 3(3): p. 199-204.
12. Jackson, A.L., et al., Position-specific chemical modification of siRNAs reduces "off-target" transcript silencing. Rna, 2006.
13. Grimm, D., et al., Fatality in mice due to oversaturation of cellular microRNA/short hairpin RNA pathways. Nature, 2006. 441(7092): p. 537-41.