Structure-guided Evolution of Ever More Potent and Specific RNAi Triggers

As the RNAi pathway is being rapidly dissected structurally, RNAi triggers continue to evolve rapidly both in terms of safety and efficacy. The 5’ end of the guide strand of an siRNA has received particular interest in that regard, and a paper in Nature now provides structural support for why human Argonaute 2, the important Slicer enzyme that cuts target mRNAs, preferentially accommodates ‘Us’ (uracil) and ‘As’ (adenine) of the genetic alphabet in its 5’ guide binding pocket (Frank et al.: Structural basis for 5'-nucleotide base-specific recognition of guide RNA by human AGO2).

Humans contain 4 related Argonaute proteins, Ago1-4, but only Ago2 can function in RNAi gene silencing. Not only will good incorporation efficiencies of RNAi triggers in Ago2 enhance gene silencing, loading less into Ago1, 3, 4 which only contribute to microRNA off-targeting without adding to RNAi efficacy should also improve the safety of RNAi Therapeutics. In fact, they may even compete with loaded Ago2 for target mRNAs and thus inhibit RNAi gene silencing. Similarly, because the RNAi function is pretty much dispensable in most cell types and developmental stages, burdening only Ago2 with the RNAi trigger should also reduce the risk of saturating the microRNA gene silencing functions of cells as Ago1, 3, and 4 should be able to cover for Ago2’s contribution in microRNA gene silencing through genetic redundancy.

Because of the structural relatedness of Argonautes, however, many RNAi triggers typically end up to some degree in all 4. In theory, however, Ago1-4 will differ in molecular detail and this provides the rationale for attempting to design RNAi triggers in a way (for example through chemistry, the nature of the double-strand RNA structure, or directing them into certain processing pathway) that the guide strand of an RNAi trigger gets preferentially loaded into Ago2. The 5’ end is of particular interest here because this is where the guide strand exhibits most ‘personality’ by exposing not only the relatively anonymous sugar phosphate backbone, but the entire nucleotide including its 5’ modification, typically a phosphate, to the Argonaute proteins.

It is therefore not surprising that, as indicated also by the patent literature, companies in the space are working on 5’ strategies that at the moment are particularly aimed at increasing the affinity of the guide strand towards Ago2. The message from the present Nature paper is apparently quite simple in that it suggests uracils and, a little bit behind, adenines to give you the best affinities. This is also consistent with the observation that the majority of microRNAs, the naturally occurring small silencing RNAs, also bear a uracil at their 5’ ends. An important technical advance in that paper is that these conclusions were drawn from studying the relevant domain (‘MID domain’) of human Ago2, including its crystal structure, while most of the earlier structure-based analyses have made use of more distantly related bacterial Argonaute proteins.

Related observations have been reported last year, when it was found that replacing the 5’ uracil of the microRNA let-7 with another nucleotide dramatically reduced its RNAi activity (Felice et al: The 5' terminal uracil of let-7a is critical for the recruitment of mRNA to Argonaute 2). Interestingly, the effect did not seem to be so much on the small RNA by Ago2, but rather at the target mRNA recognition step. It should be noted here that the present Nature studies only looked at the structure and affinities of single nucleotides, not the entire guide strand, with Ago2.

For the future, I would hope, for the noted potential for enhanced efficacy and especially specificity to see more of these studies, particularly those that systematically investigate the effect of the 5’ nucleotide on the relative incorporation efficiencies of silencing RNAs into the 4 human Argonautes. These would complement other studies, including one to which I contributed, that look at the secondary structure of the RNAi trigger with respect to the various Argonaute incorporations. A nice precedence of differential Argonaute loading based on the nature of the 5' nucleotide already exists in plants where it has been found that the mere identity of the 5’ nucleotide determines the sorting of a small silencing RNA into the various Argonautes.

Whether a 5’ uracil has discrimination effects in addition to the affinity benefit remains to be seen. However, the results already suggest that since the 5’ end of a small silencing RNA does not directly interact with the target mRNA anyway, it might be worth thinking about siRNA screening libraries that all start with a uracil. Of potentially even more immediate importance, DNA-directed RNAi promoter strategies ought to be re-considered as many of these still employ minimal hairpin RNAs driven from Pol-III promoters in which the guide strand is found at the 5’ end of the hairpin and that initiate with a non-uracil residue.

If you are a non-scientist and still reading this, you are doing pretty well. The take-home message is that while very potent, picomolar RNAi triggers can already been routinely discovered, RNAi trigger design is still an area with lots of potential for further improvements. In 5-10 years I can imagine RNAi Therapeutics to have matured to an extent where structurally highly adapted RNAi triggers effect potent gene silencing with minimal off-targeting activity.

Argonaute Gymnastics May Guide siRNA Design (Journal Club)

Argonautes are at the heart of all RNAi-related small RNA-regulated processes. A structural understanding of how they load, activate, and utilize siRNAs to seek out and repress target mRNAs would therefore prove invaluable for the rational design of potentially ever more potent and specific siRNAs. The high-impact journal Nature appears to agree, as it has been publishing at an astonishing frequency collaborative work from the Patel and Tuschl labs at Rockefeller on beautiful structures of Argonautes at the various stages of their catalytic cycle.

In the latest publication in this series, they catch Argonaute with the activated guide strand bound to first a short (12 nucleotides), and then an elongated target RNA (15 nucleotides and longer). What is remarkable is that the enzyme undergoes a substantial structural rearrangement when binding to the longer target, most notably accompanied by the popping out of the guide strand 3’ end from the PAZ pocket where it usually sits in the single-stranded form. This is thought to reflect a 2-state mode of binding fully complementary target RNAs and may differ in terms of structural re-arrangement to what happens in the setting of microRNA target recognition which is the cause for siRNA off-targeting. What is always neat to see in those structures is how they rationalize a lot of preceding molecular biology and bioinformatics work. In this case, this includes the importance of the ‘seed’ nucleotides 2-8, the sensitivity of RNAi target cleavage to mismatches and modifications, and the relative unimportance of the 3’ terminal guide bases 17-21.

The structure may help to expand on work that shows that it is possible to design small RNAs for which target cleavage and translational repression can be functionally separated. Previous work by Dharmacon (now part of Thermo Fischer) for example showed that methylation at position 2 of the guide strand abrogates its ability to translationally repress targets, while leaving cleavage intact. While those siRNAs should be highly specific, the somewhat slow adoption of this strategy indicates to me that more work needs to be done so that similar strategies are more generalizable. Based on the two-state model and present structure, such modification-induced distortions may critically affect seed-binding energy on which translational repression is so much dependent, while the conformational shift that accompanies the recognition of fully complementary on-targets, may lock the Argonaute-guide-target complex into a stable cleavage-competent conformation. The structure may help define universally applicable modification strategies that are targeted towards the 5’ end of the guide strand.

The work also demonstrates how target cleavage is sensitive to modifications around the cleavage site, and finds certain positions better tolerate them than others. This may prove to be important for aiding siRNA activation which is thought to primarily function via cleavage of the passenger strand. Finally, the series of Argonaute structures with bound guide strand provided some nice snapshots of how the 5’ end is nestled in the so-called MID-domain of the enzyme. It is not far-fetched to believe that by optimizing this interaction (e.g. 5’ end modification, nature of the base itself etc), the capture and longevity of the guide strand in the Argonaute protein can be increased with possibly increased duration of activity of the activated enzyme complex, and in the case of single-stranded RNAi, improved loading, too. Finally, this type of structural work is also highly relevant for research into how to reconcile the need for modifications for increasing siRNA stability and avoiding innate immune responses with RNAi activity.

The next major structural insight that should guide design strategies for therapeutic RNAi will be from the elucidation of mammalian Argonaute structures (humans have 4 Argonautes), as the present work focused on archaebacterial Argonautes due to the relative ease of working with them. These, however, will differ in several important aspects from mammalian Argonautes, for example in that they utilize DNA as guides and maybe even as their natural targets. Nevertheless, with such healthy progress on the molecular understanding of RNAi, the efficiency of the RNAi Therapeutics drug development platform will only increase, and it is fun to imagine what it could look like in 10 years, at which point we may be able to watch a movie on the life of an Argonaute protein by the Patel lab.

Day 2 and 3 of Keystone RNAi Conference

The Keystone conference kept me kind of busy, so I did not find the time to summarize what’s been happening here in the mountains up Vancouver, so here only a general impression on where the field seems to be heading and some highlights from the abstracts.

David Bartel (MIT) started off the second day of the conference. Well known for his groundbreaking work on microRNA discovery and microRNA target prediction work, he usually has two or three biological short stories to share that are well done and insightful. Same this time. One had to do with the technical challenge of assessing the genome-wide impact on protein expression by a given microRNA. While transcriptional profiling is a well-established and affordable method to measure changes in mRNA levels (RNA that gets translated into proteins), because microRNAs were initially believed to regulate gene expression largely by inhibiting translation without changing RNA levels, a lot of the real targets (according to Bartel’s own studies about 100-200 per microRNA) may have been missed by just looking at transcript levels (note: siRNAs largely regulate their genes by destroying the mRNA, so mRNA quantitation is a reasonable way of measuring target knockdown).

Like Klaus Rajewski (Berlin) who presented tonight, Bartel’s group compared the transcriptional impact of a microRNA by gene expression microarray and compared that to a less well known method of quantitatively measuring genome-wide changes in protein levels, SILAC. SILAC involves the incubation of cells in the presence of light (e.g. control) versus heavy (e.g. treatment) isotope media. When added at the time of the microRNA addition, newly synthesized proteins will incorporate these isotopes. If an mRNA for a given protein had been targeted by a microRNA less of that protein would be made. Cell lysates are then prepared and mixed (control and experiment) and the entire protein content analyzed by mass-spectrometry. Although the peptide signature of the mass-spec for a given protein is essentially unchanged, the incorporation of the heavy isotope would cause the profile to slightly shift for the sample grown in the heavy media, so both treatments can be compared with the relative intensity of the corresponding peptide signals indicative of microRNA regulation (say for a microRNA target the peak would be only 60% of control, meaning a 40% knockdown).

After showing that most of the changes can be bioinformatically accounted for by microRNA regulation, thus validating the technique, both Bartel and Rajewski essentially came to the same conclusions. One is that most regulated genes change both in mRNA and protein content, with the translational inhibition only accounting for a small part of microRNA regulation. This is important since this adds to the validity of assessing off-targeting by transcriptional profiling. The other good news is that most of the changes by microRNAs, the mechanism that largely accounts for RNAi Therapeutic off-targeting, are rather subtle, less than 2-fold in most cases, thereby lowering the side-effect risk through off-targeting.

Unlike previous RNAi Keystone conferences that placed more emphasis on basic principles and mechanisms, the first 3 days were dominated by high-throughput technologies. Even more so than SILAC, deep-sequencing of small RNAs cannot be avoided. It is now possible to obtain millions of sequence reads from a single sequencing reaction, mostly to discover new populations of small RNAs, but also with a trend towards quantitating them by counting the sequence reads.

Other presentations made use of more ancient transcript analysis method, the microarray, but that does not mean with less biological insight. Tom Gingeras from Affymetrix, e.g. looked at the complexity of RNAs found in human cells by microarray and cloning, and one of his observations was that a given gene locus, including those coding for proteins, often harbor multiple overlapping transcripts. That poses the challenge that even though a gene may have been well validated as an RNAi Therapeutic target, one has to make sure that the right mRNA(s) is targeted to avoid unpleasant surprises. This may be even more so important when screening the genome with siRNAs for gene discovery purposes.

Dinshaw Patel (New York) reported on the molecular structure of an Argonaute protein in archaebacteria. This work shows how the guide strand nucleic acid is bound and also suggests how it may recognize and bind its targets. This may add to our understanding of what makes for an efficient siRNA and offers the prospect for thermodynamically optimized siRNA designs. It may also inform us what types of siRNA modifications may be optimal. For example, many of the protein-guide strand interactions are mediated by Argonaute amino acid side-chains making contact in a non-sequence specific manner to the phosphate backbone of the guide, and one would probably want to avoid disturbing this interaction too much. The structure also suggests that there is quite some spatial flexibility in the 3’ end of the guide RNA which may explain why microRNAs normally tolerate 3’ mismatches with their target RNA. How about then siRNAs which are bulky at their 3’ ends such that the spatial flexibility was reduced to enforce target specificity? It certainly will be exciting to watch the impact of RNAi structural biology on the design of future siRNAs.

Similar improvements in efficacy and specificity may come from studying the function of Argonautes in humans (Argonautes are the effector proteins of RNAi). There are 4 human Argonaute proteins, with Ago-2 being the one that efficiently cleaves target mRNAs using siRNAs, whereas the function of Ago1, 3, and 4 is less well known. Some believe that all four Agos may get loaded with a given siRNA and this may very well affect knockdown efficiency and influence off-targeting. Joel Belasco e.g. proposed that the differential expression pattern of the 4 Agos in the various tissues may affect just that. Consistent with that, Dirk Grimm (Heidelberg) found that the Argonautes compete with each other for hairpin-derived small RNAs with the relative level of Ago2 determining knockdown efficiency. One idea therefore would be to design siRNAs that specifically get incorporated into Ago2 while avoiding the other Agos. The molecular structure of all four human Agos would certainly be invaluable in that effort.

Some of you may be wondering whether transcriptional gene silencing (TGS, adding an siRNA to change the chromatin structure so that the resident gene is silenced at the transcriptional level, that is less RNA produced rather than destroying it by RNAi once produced) occurs in mammals and whether it may have therapeutic utility. One problem this aspect of RNAi has suffered from was poor reproducibility and controversy around some of the studies. A particularly interesting talk was therefore given by Danesh Moazed from Harvard as his studies offered the possibility that these discrepancies may have been due to differences in the chromatin context of the targeted genes, and some human gene promoters may therefore be more amenable to TGS than others. I probably would not rush investing into therapeutic TGS just yet (there is no company based on TGS that I am aware of), but I would certainly not want to write it off either. Adding to the credibility of siRNA-induced TGS in humans, a poster by New England Biolabs, a well respected vendor of reliable molecular biology reagents, reported on the identification of efficient promoter-silencing dsRNAs.

The program of the next two days of the meeting will have more on RNAi Therapeutics, but there have been some abstracts readers here may find of interest (I had to miss those, so just the highlights here from the abstract book). One was a study by Alnylam and USAMRIID where they targeted the Ebola virus (Protiva had a similar publication with the US Army in 2006) using liposomally (probably SNALP) formulated siRNAs. Encouragingly, this formulation was able to protect both mice and guinea pigs from an otherwise lethal dose of virus- in an apparently target sequence-specific manner. Moreover, viral titers were reduced when RNAi was administered not only before viral challenge, but also after. Of course, I would have loved to ask about the toxicity profile of these treatments, but hopefully this will be addressed in one of the upcoming oral presentations of this meeting.

I haven’t had a chance to read the latest miR-122 inhibition study in Nature using Santaris’ LNA technology, but a poster abstract by Santaris described the long-lasting (up to 90 days) inhibition of miR-122 with the expected LDL cholesterol lowering effects, all of this at reasonable doses of around 3mg/kg. LNAs are certainly a promising microRNA-inhibition technology and an IPO should not be too far off. Not to be outdone, Regulus Therapeutics described an improvement of their original antisense chemistry for miR-122 inhibition with an apparently 8-fold increase in potency. I assume this must have been through a phenotypic read-out (e.g. cholesterol lowering) since they now find that this chemistry inhibits, but does not destroy the microRNA.

A lesser known method of inducing RNAi developed by Cequent Pharmaceuticals in Boston, hairpins expressed in E. coli (transkingdom RNAi/tkRNAi), is slowly getting ready to enter the clinic within the next year. In their initial program, they will orally administer tkRNAi bacteria targeting beta-catenin for the treatment of familial adenomatous polyposis. IND-enabling studies are under way.

Lastly, a poster from Rosetta Genomics illustrated the potential of microRNA diagnostics beyond cancer by detecting changes of microRNAs in the body fluid that reflected a disease state.

RNAi-related Phenomena and their Relevance for RNAi Therapeutics

Following the discovery of small RNAs as part of RNAi and a growing range of other classes of small RNAs, it soon became clear that RNAi is just one part of a network of small RNA-directed cellular processes. MicroRNAs which are involved in a wide range of post-transcriptional gene regulatory processes are probably best known, but small RNAs have also been found to play roles from chromatin silencing to chromosome rearrangement. The existence and importance of these functions can vary greatly between different organisms. Nevertheless within a given organisms these functions are interconnected by partly making use of overlapping sets of proteins or proteins from the same gene families.

One concern in the development of RNAi Therapeutics for human use would therefore be that small RNAs may engage unwanted small RNA-directed processes. In humans, the microRNA pathway and RNAi-related chromatin silencing are probably the main concerns. The microRNA pathway is actually the main reason for off-targeting in RNAi experiments, and I have outlined potential strategies about how to selectively engage RNAi and not the microRNA pathway in earlier posts. Silencing chromatin through the introduction of small RNAs in human cells, however, is much more controversial. Even if we assume that this pathway does exist, then according to published papers you would probably need more than one small RNA targeting the same gene. Moreover, one would likely have to introduce small RNAs targeted to the promoter region and not the transcribed gene as siRNAs do in RNAi. A possibly related interesting phenomena that was reported recently is RNA activation, or RNAa. In these examples small RNAs targeted to certain promoters actually upregulated gene activity. It is clear, however, that not any small RNA will have this effect, and again siRNAs targeted towards the transcribed genes are not likely to result in promoter activation.

Alone from the fact that there are 4 Argonaute genes and additional related Piwi genes, and new classes of small RNAs being discovered on almost a monthly basis, one can expect the RNAi-related universe to expand ever more. However, largely based on microarray studies, it seems that exogenous small RNAs almost exclusively engage the RNAi machinery and the related microRNA pathway. It is therefore a focus on the microRNA pathway that is most likely to ultimately improve the specificity of RNAi Therapeutics.