Solving a decades’ old scientific mystery brings us closer to discovering the perfect drug

Recently, my group and I published an important paper in Nature Biotechnology. It is important because it provides an answer to a question that has vexed scientists working in antisense oligonucleotide (ASO) technology for more than 20 years: “Why are some ASOs toxic?” In this paper for the first time we provide a step-by-step molecular mechanism that answers this question. Even more excitingly, we show that a simple chemical modification can ablate or dramatically reduce the toxicity while having little effect on potency. This opens new opportunities to more efficiently design even better ASOs.

Today, three main chemical classes are used for antisense therapeutics, the high affinity/high potency (for RNA) 2’methoxyl ethyl (MOE) class and the even higher affinity/higher potency constrained ethyl (cEt) and locked nucleic acids (LNA) (for review see Crooke, 2008).  If one screens a thousand 2’MOE chimeric ASOs (ASOs designed to serve as substrates for RNase H1), the vast majority will be safe, but a few will be quite toxic. In contrast, screens of cEts will yield a much higher fraction of toxic ASOs and the ASOs that are toxic are typically more toxic than toxic 2’MOE ASOs (as defined by the dose needed to result in liver toxicity. A screen of LNA ASOs will yield an even greater fraction of toxic ASOs.  (Freier and Watt, 2008) Thus, with regard to toxicity, there appears to be a relationship with the chemistry used in the 2’position of the ribose and the sequence of the ASO.

The question of which mechanisms could explain these observations has consumed more than 20 years of scientific effort and has arguably been the most controversial topic in the field. Two hypotheses to explain these observations have been proposed, both of which are unsatisfying. The so-called off-target hypothesis posits that higher affinity ASOs may bind to sites that contain a few mismatches and recruit RNase H1 to degrade the off-target RNAs. Though there is a solid rationale for this hypothesis, and documented examples, it is diffuse and does not globally address what RNAs may be reduced and by how much. Nor does it address the type of cell death or how the effects on disparate RNAs could result in similar patterns of toxicity in a wide range of cells and species. The second hypothesis is even more diffuse. It suggests that unknown interactions with unknown proteins are responsible for the toxicities.

In addition to the fact that off-target hybridization to mismatched RNA sequences and cleavage of the RNA in those duplexes has been shown (for review see Crooke et. al. 2008), there is a solid theoretical framework for the off -target hypothesis (Herschlag, 1991). Significant increases in affinity via hybridization are achieved primarily by prolonging the off rate for the ASO-RNA duplex. Thus, one could imagine that during the pre-equilibrium phase of drug action, higher affinity ASOs might bind to mismatched sites and RNase H1 could cause cleavage of the off-target RNA.  As most mRNAs are present at between 1-10 copies per cell, the hypothesis becomes even more plausible because even at equilibrium an infinitesimally small fraction of the total cellular amount of ASO is bound at cognate sites. Furthermore, the fact that there are extremely long RNA transcripts and that there are numerous repeated sequences in RNAs make this off-target hypothesis even more attractive.

In principle, a few factors and two rates determine whether this is the likely explanation. As hybridization to mismatched sites in RNAs is an obligate step, factors that determine the accessibility of such sites to ASOs are critical. The relative specificity of RNase H1 for perfect versus imperfectly matched sequences, the effects of repeat sequences and the length of RNAs on the probability of mismatch sites being cleaved by RNase H1 are also critical determinants, as is the fact that the ASOs being used are gapmers, which restricts potential RNase H1 cleavage to 8 to 10 nucleotides. The off-rate for various mismatched sites, the rate at which RNase H1 is recruited to matched and mismatched duplexes and the actual rate of cleavage by RNase H1 of RNA in a mismatched duplex are obviously critical factors as well.

As we learned more about the details of each step in the processes leading to cleavage of a target or off-target RNA by RNase H1, despite the theoretical framework supporting the hypothesis, I came to doubt that it could fully explain the toxicities via off-target RNA loss. First, we learned that human RNase H1 is an extremely precise enzyme. Its precision derives from the fact that it is, in effect, a pair of molecular calipers. The binding of the hybrid binding domain and the catalytic domain assures that only RNA-DNA duplexes are cleaved because the length and tilt of the duplex must be precisely RNA-DNA like. Then the catalytic domain tests the dimensions of the minor groove of the duplex and uses the 2’hydroxyl of the ribose to catalyze cleavage. In addition, RNase H1 displays sequence preference that reflects minor changes in the geometry of the duplex (for review see Lima et. al. 2008). Thus, human RNase H1 is very specific and acutely sensitive to the types of distortions in duplex geometry caused by mismatches. Second, we learned that the most important determinant of ASO binding to RNA is the structure of the RNA and that many potential mismatch sites are excluded because of RNA structures (for review see Crooke 2008).  

Next, Walt Lima and Tim Vickers in my group developed a model system that we called the SOD1 mini-gene system comprised of two exons and an intron from the SOD1 transcript. (Vickers, et. al. 2014; Lima, et. al. 2014). This system supported evaluations in a cell free system and in cells of ASO binding to an RNA whose structure we had mapped and then cleavage by RNase H1. We showed that cleavage at a mismatch site is possible but quite rare both in vitro and in cells. We created multiple cognate sites in both the intron and exons and showed that multiple sites enhanced activity by behaving independently to simply increase the probability of a productive ternary interaction between ASOs, RNA and RNase H1 and that once again mismatched sequences were highly disfavored. (Vickers, et. al. 2014) Finally, despite years of effort, we failed to demonstrate true off-target cleavages for toxic ASOs. Thus, I became convinced that off-target cleavage is unlikely to account for the behavior of toxic ASOs. Nevertheless, to perform the definitive experiment, we developed viable liver specific constitutive and inducible RNase H1 knock out mice (Lima, et. al. 2016). To my great surprise, when we tested toxic ASOs in the RNase H1 knock out mice, the toxicity of the ASOs was ablated or dramatically reduced. (Burel, et. al. 2016).

Despite the RNase H1 knock out data, there were far too many loose ends to ignore and we learned that phosphorothioate ASOs (PS-ASOs) bind a number of intracellular proteins, that many proteins displayed sequence and chemistry selectivities and the higher affinity 2’ modifications were also more hydrophobic and bound more proteins than the more hydrophilic 2’MOE. (for review see Crooke et. al. 2017). This led to the work in the manuscript that we just published. (Shen et. al. 2019). We demonstrate that toxic PS-ASOs interact differently with paraspeckle proteins than safe PS-ASOs and we show a step by step mechanism that accounts for the toxicities in a variety of cells, including those in the liver and kidney of the mouse. We show that toxic PS-ASOs induce a complex formed of paraspeckle proteins and RNase H1 that accumulates in the nucleolus resulting in inhibition of nucleolar pre-ribosomal synthesis and processing and leading to apoptosis.  Remarkably, we then show that substitution of a single 2’-methoxy nucleotide at the second gap position (gap-2’ modified) ablates or dramatically reduces toxicity with little to effect on potency, thus dramatically enhancing therapeutic index.

Evaluation of more than 700 pairs of toxic apposition gap-2 modified ASOs showed greater than 93% concordance and we confirmed that the various steps in the mechanism behaved consistent with predictions.  Moreover, we show that RNase H1 plays a critical role that does not involve RNA cleavage. Rather, the spacer domain of RNase H1 binds to a toxic ASO-paraspeckel protein complex resulting in nucleolar accumulation of the complex.

This paper is important for a number of reasons. First, it provides a detailed step by step, testable mechanism to explain the behaviors of toxic ASOs. It also demonstrates that there are a few toxic ASOs that may not be toxic via the nucleolar mechanism, thus allowing us to focus efforts on understanding additional mechanisms. Second, it provides an immediate solution to the problem of toxicity. This means that many very potent ASOs that would be discarded because of toxicity can now be used. Third, it should reduce the number of ASOs that need to be screened to identify potent and safe drug candidates. Fourth, along with efforts to enhance productive delivery, this work opens a new horizon for antisense medicinal chemistry focused on understanding and exploiting knowledge about PS-ASO protein interactions to enhance the performance of ASOs in the clinic. Finally, it once again demonstrates the value of basic research in advancing what still is relatively nascent technology, RNA-focused drug discovery.


  1. Crooke, S.T. (Ed) Antisense Drug Technology: Principles, Strategies, and Applications, Second Edition, CRC Press, 2008.
  2. Freier, S.M. and Watt, A.T. (2008) Basic Principles of Antisense Drug Discovery In Crooke, S.T. (Ed) Antisense Drug Technology: Principles, Strategies, and Applications, Second Edition, Chapter 5:117-41.
  3. Crooke, et. al. (2008). Mechanisms of Antisense Drug Action, an Introduction In Crooke, S.T. (Ed) Antisense Drug Technology: Principles, Strategies, and Applications, Second Edition, Chapter 1:3-40
  4. Herschlag, D. Implications of Ribozyme Kinetics for Targeting the cleavage of specific RNA molecules in vivo: More isn’t always better. Poc. Natl. Acad. Sci. USA 88:6921-5,1991
  5. Lima et. al. (2008) The RNase H Mechanism In Crooke, S.T. (Ed) Antisense Drug Technology: Principles, Strategies, and Applications, Second Edition, Chapter 2:47-74
  6. Vickers, T.A., Freier S.M., Bui, H.H., Watt, A., Crooke, S.T. Targeting of repeated sequences unique to a gene results in significant increases antisense oligonucleotide potency. PLos One 9(10):e110615, 2014. 
  7. Lima, W., Vickers, T., Nichols, J., Li, C., Crooke, S.T. Defining the factors that contribute to on-target specificity for antisense oligonucleotides. PLos One (7)9:e101752, 2014.
  8. Lima, W.F., Murray, H.M., Damle, S.S., Hart, C.E., Hung, G., De Hoyos, C. L., Liang, X.H., Crooke, S.T. Viable RNaseH1 knockout mice show RNaseH1 is essential for R loop processing, mitochondrial and liver function. Nucleic Acid Res 44(11):5299-312, 2016.
  9. Burel, S.A., Hart, C.E., Cauntay, P., Hsiao, J., Machemer, T., Katz M., Watt, A., Bui, H. H., Younis, H., Sabripour, M., Freier, S. M., Hung, G., Dan, A., Prakash, T.P., Seth, P.P., Swayze, E.E., Bennett, C.F., Crooke, S.T., Henry, S.P. Hepatotoxicity of High Affinity Gapmer Antisense Oligonucleotides is Mediated by RNase H1 Dependent Promiscuous Reduction of Very Long pre-mRNA Transcripts. Nucleic Acid Res 18;44(5):2093-109, 2016.
  10. Crooke, S.T., Wang, S., Vickers, T.A., Shen, W., Liang, X. Cellular uptake and trafficking of antisense oligonucleotides. Nature Biotech 35:230-237, 2017.
  11. Crooke, S.T. Molecular Mechanisms of Antisense Oligonucleotides. Nucleic Acid Ther. 27(2):70-77, 2017.