Journal Club: Structural Basis for Double-Stranded RNA Recognition by TLR3 and Activation

The recent Nature paper by Ambati and colleagues raised some concerns about whether non-specific inflammatory responses due to activation of TLR3 by double-stranded RNAs, the inducers of RNAi, would represent a significant, if not insurmountable obstacle towards the development of RNAi Therapeutics. While the short answer was no, since most in the field had already recognized the need to screen against the capacity of RNAi triggers to induce immune responses, determining the rules activating TLR3 and similar molecular patterns should facilitate the more efficient design of safe and potent RNAi triggers. Recent papers on TLR3 structural biology should do just that.

In an April edition in the journal Science, Liu and colleagues from the NIH report on the structure of mouse TLR3 with its double-stranded RNA substrate (note that mouse and human TLR3 are very similar). The structure suggests that a 40-50 base-pair double-strand RNA is optimal for binding by TLR3 thereby inducing TLR3 dimerization and downstream signaling. The fact that one of the dsRNA binding patches contains a number of pH-sensitive histidines further suggest why TLR3 signaling is most robust following uptake into endosome which provides for an acidic compartment. Overall, the structure supports previous observations that dsRNAs longer than what is typically used for siRNAs are better inducers of TLR3; however it does not explain well how smaller siRNAs may also induce such signaling.

An explanation for this is provided by structure-based mutagenesis studies by Pirher and colleagues from the University of Ljublijana (Slovenia). In this paper, they identify two dsRNA-binding patches within a TLR3 monomer and also find and explain why B-type helices, as typically found in DNA, only bind to one of the two binding patches and therefore fail to induce, and even competitively inhibit TLR3. This is in contrast to dsRNAs that typically assume A-type helical formation and bind both patches. With this the authors come up with a model to explain how TLR3 dimers bind to shorter double-stranded RNAs by assuming alternative conformations with ddimerization on the shorter dsRNAs being less efficient.

These findings immediately suggest various ways to avoid TLR3 signaling. The simplest would be to stay below 21 nucleotides, and it is well known that 20 nucleotide siRNAs are equally potent inducers of RNAi. Unlike the somewhat disgruntled 1st commentator following my previous blog would like to suggest, 15-21 nucleotide siRNAs as covered by one version of Kreutzer-Limmer in Europe are therefore therapeutically highly relevant. Generally, keeping it short is the probably easiest way to avoid non-specific immune responses. Next to sequence length, limited modifications of the siRNA at sites where they interact with TLR3 as shown by the structure should abolish any TLR3 activation and possibly serve to antagonize TLR3, similar to what has been found for 2’o-methylation and TLR7. Likewise, changing the helical shape at one end which may also have the added benefit in encouraging asymmetric RiSC loading may be a third strategy of circumventing TLR3 activation.

It is clear from reading the papers that much of the motivation for performing these is related to RNAi Therapeutics. These are findings that are not just theoretical in nature, but very much of practical relevance. The speed with which this progress has been achieved illustrates the vigor and the many tools brought to bear by the scientific community on making RNAi Therapeutics become a reality.

RNAi Therapeutics and the Indiscriminate Press

I was taken aback by the crude reporting on the back of a recent Nature paper about the potential immunostimulatory potential of siRNAs which may be responsible for any apparent efficacy seen with RNAi drug candidates currently in clinical development.

Needless to mention, the potential of nucleic acids in general, and double-stranded RNAs in particular, to elicit innate immune responses has long been known. In fact, the very discovery of siRNAs as triggers of gene-specific silencers in vertebrate cells was partly based on the hypothesis that such dsRNAs should not trigger an interferon response. That early RNAi Therapeutics programs, may not have sufficiently taken into account all the potential innate immune surveillance mechanisms present in the body, as at the time the field was still on a steep learning curve, should also not be too surprising, and I have pointed out before the risks of rushing RNAi Therapeutics into the clinic . But since two of RNAi’s most advanced, first generation candidates are nearing the crossing line, one probably understands why the press and critics of RNAi Therapeutics caught onto the findings the way they did.

A deceptively balanced New York Times article left us with the impression that the Nature paper marks a serious setback for the development of RNAi Therapeutics, instead of emphasizing that while RNAi is a relatively easily accessible tool in the laboratory, the study shows that the success of RNAi Therapeutics depends on companies and collaborations with the means and commitment of establishing a safe and efficacious RNAi drug development paradigm. I would also predict that a number of current clinical RNAi candidates will only poorly knock down their actual target in vivo and may elicit additional responses, although this does not exclude that as long as they are safe and show therapeutic efficacy they may still be considered approvable.

Scientists did their part in confusing the public to the extent that the outsider may be forgiven if he/she started to doubt the very existence of RNAi:

“The discovery is "paradigm shifting," said Dr. Charis Eng, chairwoman of the Genomic Medicine Institute at the Cleveland Clinic who works with siRNAs in cancer research. "Up until now, we all believed it's absolutely specific for gene X, so it prevents gene X from doing its job," she said.”
“Paradigm shifting?” “Absolutely”? Or how about the following passage:

"RNA interference does, of course, exist," said Ambati, a University Research Professor and the Dr. E. Vernon Smith & Eloise C. Smith Endowed Chair in Macular Degeneration Research. "It is just that siRNA functions differently than commonly believed -- not via RNA interference."

Huh? siRNA functions, but not via RNAi? I don’t think he referred to the microRNA pathway here, and, to be generous, it may have just been an awkward attempt to explain his findings to a lay person, but wouldn’t it be the responsibility of the reporter, interested in educating the public about an emerging biotechnology, to avoid such ambiguous and factually wrong statements by confirming with Dr. Ambati the scientific correctness of his statements?

Or how about the following headline from RNAi News (which btw I otherwise think does a fantastic job on reporting on RNAi developments- though not for free, of course):

“Study Shows All siRNAs Have Anti-Angiogenic Property Associated with Immune Response”

Sounds pretty dire and does not leave much to be hoped for.

Having criticized the press for careless reporting, it would be, however, equally wrong to brush the whole RNAi-immunostimulation issue aside as being made up by the press and shorts. That would be making it too easy. In an excellent and unusually candid review article on the potential innate immune response to siRNAs, Ian MacLachlan (CSO of Protiva, soon to be Tekmira) predicted to the T the Nature results and made the somewhat unsettling observation that when his group investigated siRNAs from a number of the early preclinical RNAi studies for AMD, cancer, and infection, they found a disturbing bias in the inherently immunostimulatory potential of “active” vs “control” siRNAs. Fortunately science does not stand still and it is encouraging to see that it has become common practice for RNAi Therapeutics studies to look at the potential non-specific immune responses.

A few words on the scientific and IP implications of the interplay between RNAi and innate immunity. At least in the case of synthetic dsRNA-triggered RNAi, it appears that, a priori, being appropriately modified, short and overhung is safest. In the cytoplasm, long dsRNAs may trigger interferon responses and blunt ends may be recognized by sensors of foreign RNA. TLR7 is encountered by dsRNAs (but also ssRNAs) delivered via endosomal uptake, and as the Protiva group has so elegantly shown this may be abrogated by simple 2’-O-methyl modifications. TLR3 was reported in the Nature study to be activated by siRNAs on the cell surface and thus would be invisible to siRNAs delivered within “fat globules” according to Merck’s RNA Therapeutics VP Alan Sachs referring to liposomal delivery technology. SiRNAs, however, would have to be over 20 nucleotides long to facilitate TLR3 dimerization necessary for signaling. It is therefore likely that dimerization would be particularly susceptible to modifications at the ends of the dsRNA, which should also be compatible with silencing activity. I’m sure that as I speak, the systematic analysis of TLR3 activation by siRNAs and the effect of modifications is ongoing, if not already largely completed by some groups.

One has to also keep in mind that the Nature findings were made in mice and the rules of innate immunity are notoriously difficult to translate into primates. Based on my own over-the-weekend literature research, it appears to me that TLR3 activation in humans may well occur in the endosome, and based on structural studies, 21 base-pair dsRNAs should hardly trigger TLR3 responses in human cells. In agreement with this, Dharmacon scientists have shown before that 27mer Dicer-substrates, but not classical Tuschl siRNAs trigger most likely a TLR3-related immune response. That does not mean that modifications to Dicer substrates should not be able to abrogate TLR3 activation, but shows that by choosing non-standard, less studied designs, often as a means to circumvent IP, new obstacles may emerge for which there will be relatively little support by the general research community.

Finally, since tonight is the night to challenge some of the little scientific inaccuracies propagated in the press and companies in the space, I would like to take a look at the press release issued by Silence Therapeutics that publications such as the TLR3 in Nature “SUPPORT SILENCE THERAPEUTICS COMBINED DEVELOPMENT APPROACH”. While it is true that in the two studies by Silence Therapeutics that were cited in the press release, immune responses were not found to be triggered by the siRNAs used therein, referring to Alan Sachs’ “fat globule” comment to support the rationale behind Silence’s lipoplex technology is somewhat misleading since according to the formulation method and drawings in the papers, siRNAs should be externally associated with the liposomes and therefore in theory be exposed to cell surface TLR3, and are not internally captured as is the case with the liposome technology used by Merck and others. Another little fact not mentioned in the press release is that the absence of cytokines was determined for 19mer dsRNAs and are therefore different from Silence’s claimed, and notably larger 23mer “AtuRNAi” molecules.

Journal Club: Study Shows TLR3 Induction by siRNAs with Anti-angiogenic Effects, Questioning Ongoing RNAi Clinical Trials for Wet AMD

RNAi is a relatively young therapeutic platform and we are rapidly learning about the gene-specific and non-specific effects of RNAi triggers in vivo. Ultimately, the value of RNAi Therapeutics lies in the gene-specific knockdown of therapeutic target genes, but it is clear that non-specific class effects may confound the analysis of pre-clinical and clinical results. This is especially true for a number of the early RNAi Therapeutics candidates that entered the clinic at a time when less was known about issues such as off-targeting, cytokine induction via TLR-7, and now also TLR3.

The study by Kleinman et al. from the University of Kentucky that appeared yesterday in the high-profile journal Nature (Nature doi 1038/Nature 06765), investigated siRNA therapy for wet age-related macular degeneration (AMD). Of note, there are at least 3 RNAi clinical trials ongoing for AMD: a phase III candidate by Opko Health, bevasiranib, involving intravitreal injection of an unmodified siRNA against VEGF; a phase II candidate by Allergan/Merck(Sirna Therapeutics) involving a chemically modified siRNA against VEGF-R1, also intravitreally injected; and last but not least a phase I, likely intravitreally injected “AtuRNAi” compound targeting a novel, non-VEGF pathway gene by Pfizer/Quark Biotech. Kleinman and colleagues showed that pretty much all of the siRNAs they injected, 2’O-methyl modified (Allergan drug) or not (Opko), suppressed laser-induced choroidal neovascularisation (CNV) in mice, a commonly used model for wet AMD, independent of whether their sequence was directed against an angiogenesis gene or not. Furthermore, such siRNAs were not taken up by the cells in the back of the eye, consistent with a lack of target gene knockdown. Through a series of elegant experiments, they showed that this non-specific antiangiogenic effect was mediated by binding of the dsRNA to the TLR3 receptor on the cell-surface of endothelial cells and the subsequent induction of IL-12 and interferon gamma, both of which alone could account for the observed CNV suppression.

At the moment, I cannot explain the discrepancy between these data and studies by the Opko (formerly Acuity) and Sirna Therapeutics groups that showed sequence-specific down-regulation of target-genes and cellular uptake of siRNAs using the same methods employed by the Kentucky group. Be that as it may, since TLR3 receptors are known to bind dsRNA and upregulate the IL-12 cytokine and interferon-gamma and it always amazed me how unformulated siRNAs may so efficiently be taken up in the eye, the conclusions of the studies appear credible. This receptor is different from the cytokine induction potential via TLR-7 which can be abrogated by chemical modifications, particularly 2’O-methyl.

Well, that’s the bad news. But there is also reason to be optimistic. Importantly, the authors showed that by conjugating a VEGF siRNA to cholesterol, the siRNAs were taken up into the cells and were able to knockdown its target gene and reduce CNV, even in mice lacking TLR3. All of this was achieved at the remarkably low amount of only 1ug administered siRNA per eye. This shows that RNAi could still be used very efficiently in a gene-specific manner to address AMD. Interestingly, a cholesterol-conjugated siRNA for VEGF-R1, the target for the Allergan/Merck drug, failed to ameliorate CNV, casting a doubt on the viability of this gene target that’s been relatively little characterized in the context of wet AMD. However, one should add that only one siRNA was tested for this and at the low 1ug dosage.

Cholesterol conjugation for siRNA delivery was first pioneered by Alnylam, and from recent presentations it appears that this is becoming an increasingly important technology, particularly with the elucidation of its uptake pathway in a recent Nature Biotech paper.

The authors further found that TLR induction was dependent on the length of the siRNA. DsRNAs of 21bp or longer induced this response, smaller ones did not. Also, it is very likely that this response to 21bp dsRNAs and longer could be abrogated by chemical modifications. Moreover, it would be interesting to speculate that since TLR3 is a dsRNA-specific binding protein on the cell surface, one may even harness this binding property for siRNA delivery if binding could be separated from TLR3 activation and demonstrated in the article for small dsRNAs.

Finally, as we learn more about these potential non-specific class effects of RNAi triggers (in this case synthetic dsRNAs, not DNA-directed RNAi), strategies can be designed to either avoid them by the judicious design of siRNAs (chemical modifications, length and structure of RNAi trigger) or even harnessed for a synergistic therapeutic effect. Although the number of supplemental figures of this paper (!) would indicate that there had been considerable resistance to the publication of this study, studies like this are extremely valuable in informing future RNAi development programs. This information can also be used to better monitor the safety of RNAi drug candidates already in clinical trials that may very well depend on such non-specific effects for therapeutic efficacy. Even for these drugs, not all hope is lost as firstly it remains to be seen whether and how these mouse studies would translate into the human setting. It is also true that many approved drugs work, but not through their anticipated mechanism of action, and some of the future RNAi drugs may be no exception to this.