Genetic inference of on-target and off-target side-effects of antipsychotic medications

Abstract

It is often difficult to ascertain whether patient-reported side-effects are caused by a drug, and if so, through which mechanism. Adverse side-effects are the primary cause of antipsychotic drug discontinuation rather than poor efficacy. Using a novel method combining genetic and drug binding affinity data, we investigated evidence of causal mechanisms for 80 reported side-effects of 6 commonly prescribed antipsychotic drugs which together target 68 receptors. We analysed publicly available drug binding affinity data and genetic association data using Mendelian randomization and genetic colocalization to devise a representative 'score' for each combination of drug, side-effect, and receptor. We show that 36 side-effects are likely caused by drug action through 30 receptors, which are mainly attributable to off-target effects (26 off-target receptors underlying 39 side-effects). This method allowed us to distinguish which reported side-effects have evidence of causality. Of individual drugs, clozapine has the largest cumulative side-effect profile (Score = 57.5, SE = 11.2), and the largest number of side-effects (n = 36). We show that two well-known side-effects for clozapine, neutropenia and weight change, are underpinned by the action of GABA and CHRM3 receptors respectively. Our novel genetic approach can map side-effects to drugs and elucidate underlying mechanisms, which could potentially inform clinical practice, drug repurposing, and pharmacological development. Further, this method can be generalized to infer the on-target and off-target effects of drugs at any stage of the drug development pipeline.

Fig 1. Flowcharts of the methodological processes for predicting drug side-effects.

Left: Flowchart of the data sources and methodological steps used to perform the analysis. Right: Directed acyclic graph illustrating how genetic factors are used to predict drug influences on traits via on- and off-target receptors. PSDP: Psychoactive Drug Screening Program; SIDER: Side-effect Resource; GWAS: Genome Wide Association Study; SNP: Single Nucleotide Polymorphism; eQTL: Expression Quantitative Trait Loci.

Fig 2. Complexities in predicting the side-effects of drugs.

Each block in this figure has specific problems, with our proposed resolutions. Not included in this figure is an additional challenge which is that comparison of predicted to side effects and real world side effects is hindered by lack of high coverage quantitative data on observed side effects or data from trials. Pharmacokinetics: Problem: Differential pharmacokinetics by drug can modify the concentration in the relevant tissue. Resolution: Use estimates of DDD to scale effects on the assumption that they capture differential pharmacokinetics. Blood-Brain Barrier: Problem: Some drugs have different rates in which they pass through the blood brain barrier. Resolution: An aspect of drug dose requires an adjustment for pKi and passing the blood-brain barrier, so this is also captured in the drug dose scaling. Canalization:Problem: Drug effects may be dampened or buffered by compensatory developmental processes. Resolution: It is not directly possible to detect or determine if canalization has occurred, but important to keep in consideration when evaluating results. Binding Affinity: Problem: There are different measures of binding affinity, some of which have significantly disparate measurement results. Resolution: Take a trimmed mean of the binding affinity measurements. Unmeasured Binding: Problem: Not all receptors are measured for binding affinity, and some receptors have suspected effects. Incorporating these receptors can be important to paint a full picture of drug effects. Resolution: We have decided to provide a default conservative binding affinity value based on other receptor binding affinities. Another solution could involve attempting to estimate based on protein-protein or molecule-protein interaction estimators. Direction of Effect: Problem: eQTL direction has been shown differ to protein expression, and some binding effects (like partial agonist) have non-linear binding relationships. Resolution: Report cumulative results without direction of effect, additional corroborating research must be considered before reporting a specific direction. Relevant Tissue: Problem: Drugs do not only interact with one tissue in the body, they can affect different systems, of which a single QTL measurement may not capture, especially when comparing using genetic colocalization. Resolution: Use different eQTL datasets to capture the different. We used both MetaBrain and eQTLGen. Rely on pleiotropic eQTL effects which correlate across different tissues.

Fig 3. Cumulative Effect Score per Drug.

A: Total Cumulative Effect Per Drug. Each side-effect score per receptor is summed per drug, showing the cumulative side-effect profile per drug, and the receptor that causes it. Red receptor name signifies on-target receptors, blue receptor name signifies suspected-target receptors, green signifies unknown ki values of the receptor. B: Breakdown of drug scores by receptors. Each side-effect score is summed per receptor, showing the cumulative side-effect profile per receptor across all side-effects. Red names signify on-target receptors, blue receptor name signifies suspected-target receptors, green signifies unknown ki values of the receptor. Receptors in red signify on-target receptors, receptors in green signify suspected-target receptors, while side-effects in black signify off-target receptors.

Fig 4. Breakdown of drug scores by side-effect.

Each receptor score is summed per side-effect, showing the cumulative side-effect score across all receptors. Side-effects in red signify evidence of bring primarily caused by on-target or suspected-target receptors (> 50% of cumulative score), while side-effects in black signify evidence of bring primarily caused by off-target receptors.

Fig 5. Breakdown of Specific Side-effects of Interest per Drug.

Each side-effect score is broken down per receptor, with direction indicating the relationship of the MR result of eQTL gene expression compared to phenotype, corrected by binding affinity direction. eQTL: Expression Quantitative Trait Loci. A: Scores of ‘Neutropenia’ per drug, broken down by receptor and direction of effect. B: Scores of ‘Weight increase’ per drug, broken down by receptor and direction of effect. C: Scores of ‘Blood Pressure’ per drug, broken down by receptor and direction of effect.