Recognizing and Addressing Critical RNAi Off-targeting Issue

In late 2016, Alnylam discontinued its alpha-1-antitrypsin program due to observed increases in liver transaminases. A year later, its HBV program was also terminated with the goal of advancing a new compound with increased targeting specificity.  

Apparently, close to half of RNAi triggers largely picked based on potency cause hepatotoxicity in rodent studies.  It's been clearer than ever that Alnylam, if not the industry as a whole needed to pay more attention to minimizing off-targeting and not just focus on potency in candidate selection.

Formal presentation of their findings of the causes underlying the toxicity  have now been published in a paper titled ‘Selection of GalNAc-conjugated siRNAs withlimited off-target-driven rat hepatotoxicity’.  

In an elegant series of experiments, the authors first sought to disprove that other steps in the delivery pathway that were possible candidates for the observed toxicity were responsible.  They then went on to show that it was indeed microRNA-type off-targeting that were at the root.  

Endolysosomal Constipation

GalNAc-conjugated RNAi triggers need to be highly stabilized to be active.  As a result, they reach high, antisense-type tissue accumulation, largely through sequestration in the endolysosomal compartment.  This has led to the concern that toxicities may arise from interfering with endolysosomal functions in the liver.

Strongly arguing against this possibility, when a toxic GalNAc-conjugated RNAi trigger with standard 2’-O-methyl and 2’-F-chemistry and partial phosphorothioation was rendered incapable of being taken up by the RISC RNAi effector complex, liver toxicity was abrogated without much change in the tissue accumulations of the respective oligos in rodents. 

Still, it will be important to further follow this risk in humans, especially when there is long-term, chronic exposure given that the present studies lasted for ‘only’ 5-9 weeks.  Also, patients with pre-existing morbidity in the liver may be particularly susceptible to adversely reacting to the accumulation of RNAi triggers in the endolysosomal compartment. 

Modification Chemistry

One concern with the high degree of chemical modification of the RNAi triggers which are employed to stabilize them in the absence of nanoparticle delivery formulations is that the modification chemistry could be toxic.  Rival Ionis Pharmaceuticals e.g. had previously claimed that 2’-F-modified antisense and RNAi constructs are genotoxic- at least in cell culture system.  Alnylam, however, had already disputed this claim and now argues that this is only the case with single-stranded, phosphorothioated formats Ionis is using.

The fact that preventing RISC uptake abrogated toxicity already exonerated chemistry and degradation intermediates as culprits since such abrogation in itself does not change overall RNAi trigger metabolism (note that RISC-incorporated RNAi trigger represents only a very minor fraction of overall tissue RNAi triggers).

To further rule out this possibility, the Alnylam scientists varied the degree of relative 2’-O-methylation and 2-F-modification to see whether this impacted toxicity in the rodent models.  There was no correlation between the type of modification and toxicity.

So at least for acute treatments, 2'-F modification does not appear to be a relevant tox issue for RNAi Therapeutics.  Based on the recent demonstration that 2'-F metabolites can be incorporated into endogenous RNA and DNA, there still remains a need to be vigilant about this modification regarding its potential to be genotoxic, i.e. promote cancer.

Competition with endogenous microRNAs

Years ago when I joined as a post-doc in the Kay lab in Stanford, colleagues of mine had found that high levels of RNAi trigger small hairpin precursors expressed from DNA templates could trigger severe toxicity in mice and that this could be traced back to these precursors interfering with normal microRNA function. Subsequently, studies to investigate the same were performed using synthetic RNAi triggers and modest competition with endogenous microRNA function was sometimes seen.

In contrast to the elegant investigations above, Janas and colleagues from Alnylam offered only circumstantial evidence to rule out competition.  In particular, they found no correlation between the degree of RISC occupation of the exogenous RNAi trigger and toxicity. 

It would have been interesting to study e.g. whether delivery of the RNAi triggers resulted in the de-repression of known targets of endogenous microRNAs to get insights into whether further investigation of this issue had merit.

MicroRNA-type Off-targeting

By contrast, the authors found strong evidence that it is microRNA-type off-targeting that is the culprit for the toxicity of many RNAi triggers in vivo.

To wit, in addition to directing the cleavage of mRNAs with near complete sequence complementarity (~20 base pairs), RNAi triggers may also participate in microRNA-type mRNA interactions based on limited sequence complementarity (mainly ~7 nucleotide seed) with more subdued silencing impact and without the canonical RNAi cleavage.   

Importantly, a reversir molecule with the function to soak up RNAi trigger-loaded RISC and consequently interfere with its silencing activity, reversed the toxicity.  Parenthetically, if competition with microRNAs were the problem, this ‘freezing’ of the RISC complexes would not have been expected to relieve the toxicity since it would not have freed RISC for loading the endogenous microRNAs.  

Supporting evidence came from seed swapping experiments where it was the seed from the toxic RNAi trigger that dictated toxicity. 

Practical solution

Since microRNA-type off-targeting does not translate well- if at all- between species (it doesn’t take much to change a microRNA target into a microRNA-refractory sequence), simply screening RNAi triggers in rats to weed out the toxic ones does not help much.

Fortunately, following a decade of studying microRNA off-targeting in RNAi, one chemical strategy in particular has risen to the top to minimize microRNA-type off-targeting.  Based on the hypothesis that the limited sequence interaction in microRNA target recognition renders it more susceptible to weakening the binding energy compared to a fully, or almost fully complementary RNAi target, incorporating energetically less favorable nucleotide residues were found to greatly minimize microRNA-type interactions while leaving RNAi on-target recognition intact.  

I  consider studies by Jackson et al. (2006) and Vaish et al (2011) to have been seminal, and long underappreciated, if not downplayed by RNAi industry leaders in this regard.

Similar to such prior efforts, Alnylam incorporated an ‘empty’ placeholder residue (glycol nucleic acid, GNA) which cannot base pair.  This indeed greatly reduced off-targeting as assessed by transcriptomics and, importantly, abrogated liver toxicity.  

In the end, the risk of clinical failure due to microRNA-type off-targeting can never be fully excluded before studying the RNAi trigger in humans.  Still, reducing clinical attrition due to this from 30-40% to more like on the order of 5% is well worth the effort.

Clinical studies with the new candidate for AAT by Alnylam will be the first direct test of this hypothesis.