Which ligand-protein activities cause adverse events? Interview with the author

Ines Smit is just about to finish her PhD at the University of Cambridge, where she analysed associations between bioactivities of ligands on protein targets and adverse drug reactions, funded by Lhasa Ltd. Her recent publication Systematic analysis of protein targets associated with adverse events of drugs from clinical trials and post-marketing reports has just appeared on bioRxiv, and we have now visited the author to explain the significance of her work.

Congratulations to your recent publication! Could you please summarize what you have done in this work and your PhD, and in which way this is relevant for drug discovery?

Ines Smit (IS): We analysed associations between reports of adverse effects of marketed drugs in humans and their in vitro bioactivity on protein targets. Previous analyses of such target-adverse effect associations have shaped how drug candidates are currently tested for potential adverse effects. More specifically, secondary pharmacology profiling, in which drug candidates are screened against a set of protein targets, is now widely used in the pharmaceutical industry. This followed the withdrawal of several marketed drugs in the 1990s due to adverse effects that were found to be related to off-target protein activities, such as fenfluramine’s association with heart valve disease related to serotonin receptor 2B receptor agonism, or the well-known association of hERG blockers to cause arrhythmia.

In our work, we compiled a dataset of drug-adverse event relationships from the Food and Drug Administration Adverse Event Reporting System (FAERS), one of the largest collections of adverse event reports in the world submitted by patients and doctors, and also included side effects derived from clinical trials from the Side Effect Resource (SIDER). We integrated this data with in vitro bioactivities, both experimental data from the ChEMBL database, and predicted bioactivities using the ligand-based target prediction tool PIDGIN. Importantly, we also took into account drug plasma concentrations, which we compiled from literature, by considering the ratio of activity on protein targets over the maximum plasma concentration a drug reaches (pX50/Cmax ratio). We then quantified the statistical associations between adverse events and protein target activities, in order to identify which of them are most strongly supported by empirical evidence, and may hence be able to explain adverse drug effects. This information can be used to guide the selection of targets to include on safety screening panels, as well as aid the interpretation of secondary pharmacology screening results in terms of relevance related to in vivo effects.

So how does this go beyond what has been known before, how does it advance the field?

IS: We quantified the associations using different measures of statistical associations, such as the positive predictive value, and, maybe most importantly, also took into account the drug plasma concentrations on a larger scale than previous studies. While detailed case studies including these types of information have been reported in literature, they have been restricted to a small number of target-adverse event combinations, and thus our work presents such results more systematically and on a larger scale. We also for the first time compared target-adverse event associations based on FAERS with those based on SIDER, and their distribution across System Organ Classes. Our work provides insight into areas with good concordance between in vitro bioactivities and in vivo adverse events observed, as well as areas with poor correspondence. This analysis can hence guide us towards aspects that need further attention, such as the lack of publicly available data for many reported safety targets, as well as the need for in vitro and in silico models that go beyond the protein interaction itself, in order to improve the prediction of in vivo effects, such as a closer consideration of functional effects, and more detailed pharmacokinetic information.

So what was the most difficult part of this work, and why?

IS: It is well known that different kinds of biases affect drug adverse event reports in general, and the FAERS database in particular. For example, if a side effect is already listed on a drug label, it is less likely to be reported to authorities. Although we applied a method to reduce potential confounding factors, such as drug indications and co-medications, it remains difficult to interpret results from this project due to the large number of biases present in the data. For example, observing that some previously reported target-adverse event were not statistically significant in our study, how can we distinguish between a true absence of effect, or a limitation or potential bias in our data? Incomplete bioactivity data (and absence of functional bioactivity data to a large extent), incomplete knowledge about organ-based pharmacokinetics, and unsolved challenges how to deal with severities of events also still need to be addressed in the future.

How can the results you have obtained be used by others?

IS: We provide the tables of quantified target-adverse event associations as supplementary data, so readers can find out which targets are most strongly related to their adverse events of interest, as suggestions for targets to screen in secondary pharmacology panels (or to build computational predictive models for). Perhaps this can help in making decisions based on off-target activities of drug candidates. We also provide the plasma concentrations for around 500 approved drugs, with reference to their original source in literature, which could be useful in other work.

Thank you for this conversation, and all the best for the next steps after your PhD!

Further reading: Systematic analysis of protein targets associated with adverse events of drugs from clinical trials and post-marketing reports, bioRxiv, https://doi.org/10.1101/2020.06.12.135939

/Andreas