Phenotypes associated with genes encoding drug targets are predictive of clinical trial side effects

Abstract

Biomedical scientists face major challenges in developing novel drugs for unmet medical needs. Only a small fraction of early drug programs progress to the market, due to safety and efficacy failures, despite extensive efforts to predict drug and target safety as early as possible using a variety of assays in vitro and in preclinical species. In principle, characterizing the effect of natural variation in the genes encoding drug targets should present a powerful alternate approach to predict not only whether a protein will be an effective drug target, but also whether a protein will be an inherently safe drug target, while avoiding the challenges of translating biology from experiments in non-human species. We have embarked on a retrospective analysis, demonstrating for the first time a statistical link between the organ systems involved in genetic syndromes of drug target genes and the organ systems in which side effects are observed clinically. Across 1,819 drugs and 21 organ system phenotype categories analyzed, drug side effects are more likely to occur in organ systems where there is genetic evidence of a link between the drug target and a phenotype involving that organ system, compared to when there is no such genetic evidence (30.0% vs 19.2%; OR = 1.80). Conversely, we find that having genetic evidence that a drug target is associated with diseases in which a certain organ system is unaffected decreases the likelihood that side effects will manifest in that organ system, relative to drug targets for which there is no published gene-phenotype information (18.5% vs 20.0%; OR = 0.90). We find a relationship between genetics and side effects even when controlling for known confounders such as drug delivery route and indication, and we find that this relationship replicates in an independent data set of adverse event reports for marketed drugs. We highlight examples where genetics of drug targets could have anticipated side effects observed during clinical trials. This result suggests that human genetic data should be routinely used to predict potential safety issues associated with novel drug targets. This may lead to selection of better targets, appropriate monitoring of putative side effects early in development, reduction of the use of preclinical animal experiments, and ultimately increased success of molecules. Furthermore, deeply phenotyping human knockouts will be critically important to understand the full spectrum of effects that a new drug may elicit.

Deep phenotyping of a healthy human HAO1 knockout informs therapeutic development for primary hyperoxaluria type 1

“…However, as the expected homozygote frequency in an outbred population would be only ~1 in 30 million individuals, assessing the health effects of loss of HAO1 has been challenging. Importantly, constraint in and of itself does not appear to be associated with likelihood of drug target success 10 but it is clear that phenotypes associated with variants in the gene encoding a given drug's target are a good predictor of its adverse side effects 14 . Thus the phenotype of a HAO1 null individual is expected to be informative about the likely safety profile of drugs targeting this enzyme.…”

“…Any data that informs understanding of the effects of lifelong complete knockdown of a target protein is valuable for the development of a potential therapeutic, both in de-risking the approach and providing potential hypotheses to optimize its development (e.g., biomarkers). Indeed, it is clear that on-target adverse side effects of drugs can often phenocopy the effects of impactful sequence variants in the gene encoding their targets 14 and therefore the lack of clinical phenotype in this HAO1 deficient individual is reassuring. Gene silencing based approaches, such as RNAi therapeutics, might especially benefit from such information due to their exquisite specificity.…”

Tissue-specific genetic features inform prediction of drug side effects in clinical trials

“…1; see Materials and Methods). We integrated the clinical trial drug side effect datasets and U.K. Biobank phenotypes by following the approach used by Nguyen et al (6) and mapped both side effects and phenotypes from GWA loci to 21 System Organ Class (SOC) terms from the Medical Dictionary for Regulatory Activities (MedDRA) structure. We then integrated this dataset with gene expression and eQTL information from the GTEx project by mapping the 21 SOC terms to the 48 tissues repre sented in the GTEx dataset (Materials and Methods and table S1).…”

“…There are a number of potentially responsible factors, including lack of GWA loci due to reduced statistical power for certain phenotypes, lack of representation for side effects that do not have related pheno types that were tested in genetic association analyses, lack of relevant tissues, and, additionally, the reporting of clinical trial side effects that were not directly caused by the relevant drugs. Fourth, the lack of granularity in the SOC term and GTEx tissue mapping methods that enable integration across datasets [described in previous studies (5,6) and also used here] may result in the misclassification of cer tain phenotypes and also greatly limits the utility of our approach to be used for individual prediction of side effects for drug target genes. Therefore, our findings should be viewed as providing proof of con cept.…”

“…The 1819 drugs were selected based on a protein target present in at least one of three databases [Pharma projects (Informa PLC)] (26,27). Drug route and drug modality were also provided and described by Nguyen et al (6). Exclusion criteria as defined by Nguyen et al (6) included drugs from oncology and combination therapy trials, drugs with common side effects, and non-smallmolecule/biologic drugs.…”