Publications

Presentations/Lectures

This page decribes the strenghts and limits of using RNA interference in biology and is the text version of a presentation I gave about a topic given by the jury of my PhD defense (see the presentation). Since the literature on RNA interference is quite extensive, it only comprises things that caught my particular interest .. 

This page addresses :

1. Introduction into RNAi

• Where

• What

• Why

2. RNAi in Biology

• Strengths

• Limits

• Overcoming limits

3. Conclusion

4. References

Let's first start with the basic concepts

RNAi stands for RNA interference. This implies that there occurs some kind of interference, .. but on whose part ? Is it RNA that interferes with another compound, it it a certain compound that interferes with RNA or is it actually RNA that interferes with RNA ? The aim of this section is to clear this out.

Although the major breakthrough on "RNA interference" or "Post-Trancriptional Gene Silencing"  as a general mechanism for gene silencing was first reported in the worm C. elegans, the phenomenon had long been known in fungi (like Neurospora crassa) as "Quelling" and in plants as "Co-suppression" or "Post-Trancriptional Gene Silencing". After the discovery in worms, researchers became interested in exploring whether this phenomenon was preserved in flies (Drosophila melanogaster) and mammalia. And suprise, surprise, RNA interference also occurred in these species ..

Quelling	PTGS Co-suppression	PTGS RNA interference	PTGS RNA interference	RNA interference

So how did researchers find out about RNA interference ? The starting assumption was that the introduction of antisense RNA would inhibit translation of mRNA complementary to the introduced asRNA.  This would  occur by hybridization of both strands, which physically prevented access of the ribosome to the targeted mRNA and in this way passively suppressed gene expression. An alternative way to suppress gene expression comprised of the recruitment of nucleases which led to the direct  destruction of mRNA.

But, when Fire and coworkers used this strategy to knockdown the unc-22 gene product, they found something bizarre.. Introduction of dsRNA, as well as single-stranded antisense RNA ànd the control sense strand could elicit gene silencing ?! (Although the dsRNA molecule appeared far more potent than the other two)
What was even more puzzling wa  that this specific gene silencing could spread from one tissue to the next and even be transmitted into the germline!

The fact that this specific silencing could spread, meant that at some point, one would require some sort of amplification mechanim, because the minuscule amounts one introduces to elicit gene silencing are not enough to dilute it out over next generations of cell or new tissues .. which means then.. that it is an active proces !

Now, how would this work ??

Drawn to the right, you see we have some sort of black box scenario. This results from what we know now : that dsRNA potently silences gene expression.

So, how can we find out more about this process ?
Let's take it one step at a time..

What do we know about the nature of the RNA that can induce this gene silencing ?

Apparantly not all types of RNA can induce gene silencing. Examples of RNA that can, are depicted in the scheme on the left.

 The RNA can originate from an exogenous source like a synthetic RNA molecule, a transgene or a virus. In addition, silencing can also be induced by endogenous RNA. In that case, directed or inverted repeats, transposons and retro-elements may all be suited for the induction of RNA interference induction (in the yellow frame endogenous RNAs that can induce silencing in mammalia, in the green frame the ones in flies).

Now since the long dsRNA could obviously not be transmitted to other tissues or the next generation, the signalling compound had to be smaller. Therefore, researchers decided to look for small RNA molecules that occurred during PTGS in plants. They found the small RNAs in the 20-30nt range and characterized them into several categories depending on the origin of their precursor. The three most well known small RNA molecules are siRNA, miRNA and piRNA, whose origin is indicated by the blue, red and yellow colours in the upper figure respectively. Here, "Genome encoded" indicates precursors that originate from coding or noncoding transcripts that exist as individual miRNA genes or tandemly arrayed clusters e.g. miR23.

So to summarise, we have our big RNA precursor, we
know that small RNA molecules are involved and that this ultimately leads to gene silencing.

So what can we expect next ?

Reason tells us that in order to go from large RNAs to small RNAs, we will probably need a cleavage of some kind and to go from small RNAs to silencing of specific genes we will probably need some homology of the small RNAs with the target genes, that will either lead to cleavage of the targeted mRNA or prevent translation by hybridising to the targeted mRNA.

In what follows I will mainly elaborate on the mechanims for RNA interference or miRNA induced silencing and refrain from discussing the piRNA induced silencing mechanism because this is poorly characterised.

For the elucidation of the mechanisms involved in the generation of siRNAs and miRNAs, the breakthrough was made in plants _again_, where a nuclease called Dicer was discovered that could produce 21-23 nucleotides with 2nt 3' overhangs. Only one strand of the resulting dsRNA molecule will then be incorporated into the RNA induced silencing complex or RISC, whereas the other strand will be degraded. The RISC complex than finds mRNA molecules that contain complementary sequences to the siRNA strand inside RISC and cleaves the mRNA molecule opposite to the middle of the siRNA sequence. Although it was previously believed that the siRNA and miRNA pathway were different pathways, nowadays researchers believe they are the same.

So therefore I will explain this pathway as well. Endogenous pri-miRNA transcripts are processed by a Dicer like enzyme called Drosha, which cleaves the precursor in a 70nt hairpin molecule with 2nt 3' overhangs. This molecules is then recognized by Dicer which cleaves small RNA molecules of the same size as siRNAs, where the passenger strand of molecule is destroyed via a bypass pathway, and the guide strand is taken up in the RISC complex. The complex would then scan for complementary targets but unlike in siRNA, it would lead to translational inhibition because of some mismatches between the miRNA and the target RNA (which inhibit binding of the ribosomes), but it would nevertheless result in gene silencing. In plants and nematodes silencing can spread. The protein that may help to spread the silencing effect is called Sid2 (what Sid1 and Sid3 do, remains unknown), which can form a channel between two adjacent cells, through which the small RNA molecule may travel.

Why would different organisms want to use RNAi?

RNAi can maintain genomic integrity by limiting the spread of transposons and the accumulation of repititive DNA. The RNAi can be used as a defense against viral infections. Since viruses often use RNA intermediates during their replication process, these may be destroyed immediately and thereby limit viral replication and spread. Finally, miRNAs can guide gene expression. Even more they were discovered in C. elegans to be important during development.

So let's now take a look at how RNA interference can be used in biology as a technique ..

In biology we use RNA interference mainly to study protein function. The advantage of using RNAi for this purpose is that we don't need to disrupt a gene about whose function we want to know more. RNAi can recognise its targets with high specificity which makes it very suitable for discriminating mutations in genes that cause diseases.

In this way, we can compare compare healthy and diseased cells with or without the use of siRNAs (or short hairpin RNA molecules) or introduce these molecules at different stages during development. We can then screen for markers of a particular cellular process and in that way find out whether the gene of interest may play a role there. The anwers we get from this study is that it gives us more information about the protein and its function on the molecular level, but in a larger context it will enable us to unravel how signalling is disturbed in disease processes (and thereby identify key-points for therapeutics).

The following scheme gives a rough indication of how we can employ RNA interference as a method in cell culture:

Since we want to knockdown our “gene-of-interest”, which differs between different research groups, we would need to devise oligonucleotides that function as siRNA specific to the gene we want to investigate. Once this is done, the siRNA has to get into the cell somehow, where we later have to show that it is active, and has silenced the gene we intended.

In the next section I will comment on each step it advantages and its limits and where possible offer solutions to compensate for the limitations.

(1) In practice we use siRNA molecules or shRNA molecules. siRNA molecules are typically 21-23 nucleotides with 2 nucleotide 3' overhangs, like they are processed by Dicer.
For shRNAs there exist two strategies: a type 1 hairpin has an almost perfect match in the base and a very small loop, type 2 hairpins may contain some mismatches and have a more extensive loop. This type of hairpin resembles the miRNA precursor and will be dealt with as an miRNA product by the cell.

The question for the use of these small molecules lies in how specifically such a molecule can recognize its target or rather what mismatches does it tolerate? 

The first example addresses strand specificity in viruses.
When one wants to reduce the viral spread, RNAi could help to degrade viral RNA of particular viral genes that are required for the propagation of the virus. For example Coxsackie virus B3 is a positively single stranded RNA virus that belongs to the enterovirus group in the picornaviridae. It causes myocarditis and currently no vaccine is available. Since these viruses replicate via the synthesis of negative ss RNA, the question is whether silencing of genes on the positive strand ends up more advantageous for the host cell compared to silencing of the negative strand?

In order to solve this question, Schuber and collegues devised two sets of reporter vectors : one containing the gene for EGFP, the other for luciferase. Therein, they cloned the viral RdRP of CoxsackieB3 virus in sense and antisense orientation and transfected this into COS7 cells together with siRNAs against RdRP. Since siRNA molecules can recogize both the antisense and sense RdRP gene, and the researchers wanted to be able to discriminate between the strands they introduced Locked Nucleic Acid (LNA) at discrete positions: LNAs on the last base pair at 5'end of the sense siRNA increases its thermodynamic stability, which turns the 5' of the antisense into a lower binding energy pair to ensure incorporation into RISC. Incorporation of LNA at position 10, 12 and 14 of the sense strand impairs its participation in mRNA cleavage thereby reducing off-target effects due to the sense strand. (The modified strand is not incorporated into RISC and hence will not cause silencing of of RdRP, and inhibition of viral propagation.) By performing plaque assays, they found out that when they targeted the positive ss viral RNA, they required 10 to 100 times less siRNA, compared to the negative strand, in order to obtain the same amount of viral inhibition.

Aside from being able to discriminate between different strands, siRNAs can apparantly also distinguish between a few nucleotides. In a second example on the specificity of siRNA/shRNA, researchers assessed whether they could silence genes that differed by one nucleotide. In this case, they wanted to distinguish between wild-type and mutated superoxide dismutase 1 (which differ by only a single nucleotide), the mutated form of which results in increased toxicity during amyotrophic lateral sclerosis.

They introduced different types of mismatches at different position in the siRNA and assessed whether the position alone could lead to different extents of gene silencing. They found that mutations in all positions, except for position 16 would be indiscriminative. In addition to the position, the nature of the mutation also seems to exert an effect on the specificity of silencing. In the table below, we see the one nucleotide difference between the WT and the disease causing mutant.

Gene	WT	Mutant	Molecule
SOD1	G	C	mRNA
	C	G	anti-sense siRNA molecule

For an siRNA molecule to be able to recognize the WT or the mutant, the complementary bases are indicated as well. Now, what seems to be so special is that only when the mismatch between the WT molecule and the antisense molecule that targets the mutant exists as a purine:purine mismatch (like here G in the WT and G in the as siRNA strand), can the siRNA discriminate between the WT and mutant. In other words, only in these circumtances can the siRNA distinguish the mutant from the WT and thereby destroy the mutant and preserve the WT.  

These examples obviously illustrates the power of RNA interference to silence the target genes.

But now we wonder : do highly specific siRNAs always recognize their target with the same precision, in other words, could it be possible that once in a while unintended genes become silenced ?

In order to find this out, Jackson and coworkers designed 16 siRNAs against the same IGF1R gene and 8 siRNAs against p38®  gene, also known as MAPK14. Next, they transfected HeLa cells with these siRNAs and assessed by microarray whether all siRNAs targeted the same genes as they were supposed to and whether siRNA treatment could also silence gene expression of noninteded targets. As we would expect, they found that each siRNA elicited gene silencing in the same way in three independent experiments, however, unexpectedly, they found only limited similarity in the gene expression profiles of different siRNA molecules that all targeted the same gene!

They next wondered whether these effects were less present at lower concentrations and thus performed a concentration and kinetics analysis. But unlike they expected, they could not get rid of the silencing of unintended targets. This indicates that the concentration of siRNA is not a major factor in off-target effects.  Next, they examined the nature of the different transcripts that were altered by siRNA treatment and found 4 groups:

• Group 1 transcripts contained the target transcript p38®  ,

• Group 3 and 4 transcripts were altered as a secondary effect due to the knockdown of p38,

• Group 2 on the other hand contained transcripts that showed homology to the siRNA molecule but were not a part of the p38 signaling pathway. Furthermore these genes were knocked down much sooner than p38, which indicates that they are probably not the result of p38 knockdown but rather due to unintended recognition of the siRNA molecules. These 9 transcripts showed homology to the core region of the siRNA or to the last 9 nucleotides from the 3' end, but only the 3' end nucleotides seemed to be involved in sequence specificity and hence were labelled as unintended off-targets.

Further analysis revealed that the minimum requirement for off-target effects with sequence homology was 11 nucleotides and that position 15 of the siRNA was the most discriminative (similar to what other groups reported later). The authors also found they could not predict which siRNA would elicit which off-targets, making this an especially nasty limit of siRNAs.

What can we do to avoid off-target effects when we want to investigate protein function after knockdown of a particular gene ?

• At the design stage we can make sure that we include the possibility for off-target effects, by searching for homology to other sequences at particular positions. Iyer and coworkers developed a stand-alone program that allows for particular requirements during the search for an effective siRNA.
  
• In addition, we can also introduce LNA or other types of modifications in the sense or antisense strand of the oligonucleotide.
  
• We can also use multiple siRNAs and assess each phenotype carefully. When we find a consensus phenotype for most siRNAs, it will be very unlikely that the effect can be attributed to off-target effects.
• We could compare our phenotypes with knockout cell lines.
  
• Finally we can try to rescue the obtained phenotype by introducing the target gene, but one that is resistant to the siRNAs

A workable vector for such a rescue approach has been developed by Hulzel and coworkers. Here, they deigned two episomal constructs that could induce gene expression from a bidirectional promoter:

1. One construct bore a luciferase and an miRNA-flanked siRNA that would target the gene of interest and thereby induce its knock-down.
2. The other construct would express either the gene of interest as a rescue for the knocked down gene  or  a mutated gene (for example in a protein binding domain or activity domain, in order to investigate the impact of such a mutation on the function of the protein), which would need to be resistant to the siRNA. In addition, this construct also expressed a fluorescent protein as a control for the presence of the plasmid.

Both constructs were responsive to doxcycyline, where doxcycline turned on the gene expression and the expression of the siRNA, whereas removal of doxycycline inhibited gene expression.
They used rRNA processing factors PES1 and WPR12 as genes-of-interest. These genes are involved in rRNA biogenesis and result in impaired processing of 32S pre-rRNA and reduced cell proliferation in H1299 cells (NSCLC cells), a phenotype that can be detected easily. Assessment of a panel of functional and non-functional mutants of WDR12 confirmed that the assay was functional and in addition it provided the isolation of new proteins involved in rRNA biogenesis. Therefore, the KD-KI approach may help to determine in a more controlled way whether a particular phenotype is due to off-target effects or to knockdown of the targeted gene.

(2) Since we now have an idea about what factors to take into account when designing oligos, we can proceed to the next step, namely the introduction of siRNAs in the cell. This part also includes a brief mention on the responses of the cell to the introduction of this foreign material.

Plants and worms are able to take up siRNAs from their environment and can transmit this siRNA signal to other cells or distribute it over their tissues and even transmit it to the next generation. This contrasts to mammalian cells that unable to spontaneously take up siRNAs. This means that we need an artifical way to introduce these compounds into the cell. We can do this with invasive techniques like micro-injection and electroporation, or we can deliver siRNAs in a more natural way : via liposomes, via cell penetrating peptides or internalizable antibodies. The fact that the receptivity of each cell type towards these molecules determines the efficiency of silencing, means that  silencing effect may very between transfected cells.

In contrast to plants and worms, mammalia lack the amplification and propagation responses. This implies that when we introduce siRNAs into mammalian cells the RNAi effect only lasts temporarily and in order to overcome this caveat, we may introduce shRNA via viral vectors or the construction of stable cell lines expressing the shRNA.

An example of such a methodology was recently provided by Dann and coworkers. They knocked down the gene for Dazl, an RNA-binding protein required for fertility. In mice, knock-out of Dazl results in male and female sterility. When using the shRNA vector (indicated in the figure on the left), they found male sterility, which nicely indicates how RNAi can indicate the involvement of a particular gene product in physiological function.
Figure from PNAS 2006;103(30):11246. Dann et al.

However, the introduction of siRNA is a quite stressfull event for the cell, and numerous reports have described that shRNA and siRNA can elicit the interferon stress response in cells, which ultimately leads to apoptosis or cell death. This reaction is vertebrate-specific and is initiated when dsRNA, blunt ended ds or ss siRNAs are introduced into the cell or when the cell can detect 2'OH uridines. In contrast to the off-targeting effects, this response is concentration-dependent, and occurs even when the molecule shows no similarity to the gene-of-interest.

The cellular stress response

The figure on the right (from "RNA 2007;13:431. De Paula et al.") provides an overview of the different stress pathways that can all lead to inhibition of protein synthesis and consequently cell death upond the introduction of foreign material. The most famous pathway is the PKR pathway where dsRNA activates PKR which then phosphorylates eIF-2alpha which then leads to inhibition of protein synthesis. dsRNA can also trigger oligoadenylate synthases which then activates RNAseL and causes unspecific degradation of mRNA. Blunt ended siRNA can be recognized by Retinoic-acid Inducible Gene 1(RIG-1) which unwinds the molecule and thereby activate NF·B and Interferon Responsive Factor 3 (IRF). These compounds induce the synthesis of inflammatory cytokines and thereby the death of the cell. Similarly Toll-like receptors (TLRs) in the endosomal vesicle, can also induce NF·B and IRFs.

How can we avoid triggering the stress response of the cell?

For siRNAs, we can :

• avoid A/U nucleotides at the 5' end,
• design siRNAs so that they won't hybridize with sites for mRNA-binding proteins
• assure that :   
• the molecule contains 2nt 3' ends for more effective cleavage
• hybridization between the siRNA and the target mRNA results in an A-form helix.
• the G/C content lies between36-56%
• the siRNA has the most effective size : 21-23nucleotides or 25-29 nucleotides.

(3) When we have introduced our siRNA into the cell, we can next concern ourselves with the activity of the RNAi machinery. Besides the introduction of a single siRNA or shRNA one can also knock-down multiple genes at once. So let's see what this possibility has in store for us.

This approach may help us target genes that encode redundant proteins at once, an approach considered by Zhu and colleagues. They constructed a vector with a replacable promoter and two miRNA flanking sites (for long-term expression and transcription from a Pol II promoter) in between which one can insert a shRNA. 

In a first stage, they tested whether they could downregulate 3 genes in single, double and triple KD constructs for Arrestin2, GPCR kinase 2 and G-protein beta-2; and in a next step they created stable cell lines of RAW macrophages to KD gene pairs of Arrestin 2/3, Grk2/5 and PKAcalpha/beta. In these cell lines they could KD the proteins as expected.

Since they previously performed a microarray experiment to screen for cAMP dependent mRNAs, they selected four genes for assessment in the PKAc® /¯ KD cell line.These genes were :

• Nr4a1 and Nr4a2,which are members of the nuclear orphan receptors regulated by cAMP;

• Ctla2® and Ctla2¯, which are cysteine protease inhibitors originally isolated in cytotoxic T-lymphocytes, whose transcriptional regulation has not been investigated in this context.
  

KD of either PKAc®  or PKAc¯ did not result in a decreased response after administration of cAMP analogues, however double KD markedly decreased the transcripts of these four genes. This indicates that KD of redundant genes may indeed provide a clearer picture of a proteins function or involvement.

After being able to target redundant proteins at once, the question is whether one can KD genes that signal in the same pathways to estimate the role of these genes in a particular processes targeted by the pathway. For this Sahin and coworkers knocked down the ErbB2 receptor tyrosine kinase and the protein kinases Akt and MEK1. They first determined at which concentrations all three genes would be moste effectively downregulated without causing too much cytotoxicity. This turned out to be the lowest concentration. Next, they assessed whether other genes that would not signal in the same pathway were affected. They found that although RNA levels could display high fluctuations, these were not translated onto the protein level when all three genes were knocked down.

To determine whether the knock-down of different combinations of these gene would result in differences in cellular function, they transfected cells with, single, double and triple KD shRNAs and assessed their invasion potential. Their results confirm that Akt, one of the main players during metatasis, is mainly involved in invasion and the ERK1/2 pathway in proliferation. ErbB2 KD does not reduce the invasion capacity and suggests that Akt can be activated via other means. A double KD of ErbB2 and MEK1 implies that other receports can target Akt which leads to a reduced cell invasion. The triple KD does not completely eliminate invasion potential thereby suggesting that parallele mechanisms can intervene to attenuate the invasion potential.

Although off-targeting remains a problem in these experiments, things are even more complicated with multiple siRNAs or shRNAs: RISC can only bind a limited amount of siRNAs, so the use of high concentrations of siRNAs results in many unbound molecules in the cytoplasm and the initiation of the cellular stress response. Even more, during increased siRNA concenrations, only the “fittest” siRNAs are incorporated into RISC leaving the less potent molecules wandering in the cytoplasm. shRNAs can cause saturation because they compete with the endogenous miRNA pathway for processing power. Such competition can induce toxicity as the next example illustrates.

Grimm and coworkers introduced either liver-targeted luciferase or human ®1 anti-trypsin and specific shRNA in mice. This introduction caused weight loss and death when high concentrations were administered. They found that this toxicity was accompanied by a reduction in the amount of liver miR-122 and let7a and remained independent of liver injury induced by chemicals or surgery. shRNA function was decreased in the presence of shRNA or miRNAs and indicated that shRNA indeed competes with the miRNA machinery. One bottleneck for processing seems to be exportin 5, the nuclear exporter of pre-miRNA, since overexpression of this protein, increased silencing again in presence and absence of shRNA/miRNA competitors.

The last example of how siRNAs or shRNAs may be used is at an even more increased level of complication, namely the genome-wide RNA interference. Such an approach may yield a lot of data and understanding but it requires specialized equipment(robotics) and very carefull planning.

An example of such a genome-wide analysis concerns a cell morphology study where two different Drosophila cell lines where treated with a dsRNA library and microscopically checked for parameters like cell number, size, shape, viability, actin filaments, microtubules and DNA. In addition, a co-RNAi screen was performed with PTEN to assess its role in cell shape. Knock-down of approximately 100 genes resulted in a visible effect amongst which several modifiers for PTEN, a phosphatase that dephosphorylates PIP3 (and a functional antagonist of PI3K).

This indicates that when performed carefully, genome-wide RNAi screens may yield valuable data on protein function and their roles in cellular processes, providing that all information is carefully validated. Unfortunately, today there exists too little overlap (stochastic and biological errors,off-target effects..) between the results obtained in different genome wide RNAi studies to incorporate all data in the same databases.  

Conclusion :

Using RNA interference in biology provide very usefull. However, despite it strenghts (specificity, multiple targets and use in transgenic animals), its use also includes several limitations (off-target effects, cellular stress responses, saturation.

Ideally, one can -at least- partially account for off-targets by carefull design of oligos. By including at least one marker for the interferon stress response it should be possible to discern whether or not one elicits a such a response. In addition, one can use several siRNAs against the same target gene, to partially determine which genes are unwelcomed targets and to determine phenotypes, it may help to use a knock-down rescue model to ensure that the phenotype one sees does not result from off-target effects, saturation effects or stress responses.

References :
Current Opinion in Molecular Therapeutics;2003:5(3):217 Paddison and Hannon.
Current Opinion in Molecular Therapeutics;2006:9(3):248. Svoboda.
Nature Reviews Genetics 2007;8:884. Chapman et al.
Nature 1998;391(6669):806. Fire et al.
Nature 2004;431:338. Mello and Conte.
Nature 2004;431:343.Meister and Tuschl.
Nature 2004;431:371. Hannon and Rossi.
Nature 2004;431:364.Lippman and Martienssen.
Nature 2004;431:350. Ambros.
Antiviral Research 2007;73:197. Schubert et al.
PloS Genetics 2006;2(9):e140. Schwarz et al.
Nature Biotechnology 2003;21(6):635. Jackson et al.
RNA 2007;13:431. De Paula et al.
Cold Spring Harbor Symposia on Quantitative Biology LXXL, 2007. Prevot and Perrimon.
BMC Molecular Biology 2007,8:98. Zhu et al.
PNAS 2007;104(16):6579. Sahin et al.
PNAS 2006;103(30):11246. Dann et al.
Nucleic Acids Research 2007;35(3):e17. Hülzel et al.
J.Biology 2003;2:27. Kiger et al.
Computer Methods and Programs in Biomedicine 2007;85:203. Iyer et al.
Nature 2006;441:537. Grimm et al.
Nucleic Acids Research 2005;33(1):439. Elmen et al.
PNAS 2006;103(15):5953. Dykxhoorn et al.
RNA 2006;12:1197. Jackson et al