When antisense oligonucleotides are dosed systemically in rodents, injury to the liver is sometimes seen. The liver is one of the organs that accumulate the most oligonucleotides, and this may be part of the reason. However, some oligonucleotides can accumulate to very high concentrations in the liver without any hepatotoxic reactions, whereas others elicit hepatotoxicity at much lower levels. This indicates that each oligonucleotide has an inherent hepatotoxic potential; the lower this potential, the higher the dose needed to elicit a hepatotoxic reaction, and vice versa.
p>When considering the dose at which any drug starts causing toxicity, this of course has to be related to the dose at which it has beneficial therapeutic
effect. Still, it is uncommon for drug developers to continue development with products that elicit overt hepatic toxicity early in the animal testing, and oligonucleotides with high hepatotoxic potential as evaluated in rodents are usually not progressed to clinical testing in humans. It is therefore very important to
Below is shown a histogram stratifying oligonucleotides according to their hepatotoxic potential as measured in mice.
For the past three years, scientists from COAT and Santaris Pharma have been exploring ways to decompose the chemical structure of RNAse H-recruiting oligonucleotides in a manner that allows the hepatotoxic potential to be predicted by a computer algorithm. Recently, we published a study in Nucleic Acid Therapeutics, the official journal of the Oligonucleotide Therapeutics Society (Hagedorn et al., 2013), which shows that this is indeed possible. A press release from Mary Ann Liebert, Inc., publishers can be found here. By decomposing the oligonucleotide sequence and modification pattern into dinucleotide counts, we demonstrate that a random forests classifier distinguish between oligonucleotides with high- and low hepatotoxic potential with more than 80% accuracy.
Since tens of thousands of RNAse H-recruiting oligonucleotides can be designed against an RNA target (see previous blog entry) such a hepatotoxicity predictor can for example be applied at the design phase, to ensure that most of the oligonucleotides that are actually synthesized and evaluated in vitro and in vivo, have a low hepatotoxic potential. Using the classifier, we furthermore demonstrate how an oligonucleotide with otherwise high hepatotoxic potential can be efficiently redesigned to abate hepatotoxic potential. So, alternatively, the hepatoxicity predictor can be used for lead optimization once a highly potent oligonucleotide has been identified.
As seen from the example shown in the figure above, the study supports the view that the hepatotoxicity profile of an oligonucleotide is unique to the specific oligonucleotide compound, and slight alterations may result in a profoundly different hepatotoxicity profile. Since the training and validation of the classifier was based on screening data for 236 locked nucleic acid (LNA)- modified, RNAse H-recruiting, antisense oligonucleotides, it will be interesting to see whether the same kind of predictive performance can be achieved with other types of chemical modifications or therapeutic mechanisms.
This work will also be presented at 9th Annual Meeting of the Oligonucleotide Therapeutics Society in Naples, Italy (October 5th-8th, 2013) as a short talk (Session IV) and poster.