Success In Finding SAR For Seizure Inducing Compounds

Seizures caused by central nervous system (CNS) toxicity is one of the most serious issues in drug development, leading to project delays and terminations. Approaches to reducing these risks would be very welcome. Although these CNS risks can be identified in animal studies, it is highly desirable to move to in vitro assays. Ideally this would allow for the prediction of seizure risks, or at least identify possible structural flags to prioritise compounds for testing. This is both for speed, cost and to continue to reduce the number of animals used in drug discovery and development.

Apconix have been leading the development in new approach methodologies by assessing the effects of known seizurogenic compounds on the electrical activity of stem cell derived neurons. This is followed by building the causal chain back to identifying which ion channels these compounds have an effect on. With potential causal ion channels identified, it is then possible to:
• build panels for screening,
• start to assay compounds,
• start the process of identifying if there are specific pharmacophores associated with seizures.

MedChemica worked with Apconix, first to identify potential compounds for testing and then to see if an early signal could be found in the chemical structures as a warning for chemists.

In cheminformatics terms this was a relatively tough challenge. We had a limited set of known actives and the potential for acquisition from the Enamine REAL database of billions of make-able compounds. With a limited compound acquisition and testing budget to work in, we set ourselves the task of picking less than 100 compounds to test across 4 targets. It is essentially a virtual screening problem, but we wanted more than just actives, the aim was to generate some limited SAR across multiple ion channels that could be shared with chemists and understood by them.

It is not enough just to pick compounds “near” the actives by some similarity metric, any SAR we found should be understandable and then testable with further compounds. Of course a concern was that with only a very small testing set, either all the compounds would be inactive – the most likely outcome, or the SAR would be completely flat with no variability, just weak binders. For this problem, we chose to use our implementation of the idea of pharmacophore pairs. In this approach, molecules are described as having a pair of pharmacophore features (hydrophobic group; hydrogen bond donor; hydrogen bond acceptors; aromatic ring; aliphatic ring; halogen bond donor and acceptor; basic group; acidic group) separated by a topological distance through the molecule, i.e. a number of bonds.

There is some subtlety in our implementation, having spent significant time in honing the pharmacophore definitions, particularly of the acids, bases, hydrogen bond donors and acceptors to be more specific than the generally used open-source definitions. It was extremely gratifying to see that these descriptors proved to be successful in both finding active compounds (in 3 / 4 channels finding compounds with IC50’s < 10µM from query molecules of similar potency) but also seeing SAR in the assay data and best of all achieving statistical significance for pharmacophores in the α4β2 nicotinic receptor, KV2.1 and α4β2 nicotinic receptors.

Only small numbers of compounds were tested, so the results are very preliminary. The experienced pragmatist knows that such small data sets should not be relied on too much and QSAR models often need 100’s of IC50 measurements to make them robust. At a very early stage, the data shows enough signal to warrant further investigation. For a tiny virtual screening campaign, we were very pleased with how it turned out. If we consider how the understanding of cardiac liabilities has altered in the last thirty years, most chemists would use both an understanding of pharmacophores that contain a hERG risk and routine testing to avoid cardiac toxicity in their development compounds. By analogy, we can imagine with a similar set of screens and enough data we could reduce the CNS ion channel driven seizure risk, reduce animal testing and prevent projects being delayed or terminated earlier.

For now, using the seizure ion channel panel is one recommendation we have for our clients. If compounds get into the CNS, it is worth looking at in late lead optimisation, because there is now a better a way forward on seizure risk.

Read the full paper here: https://academic.oup.com/toxsci/advance-article/doi/10.1093/toxsci/kfag094/8747637?login=false

For more information about our consulting services, click here: https://www.medchemica.com/consulting/