A UCLA study has outlined a new framework that researchers say would improve predictive power of genetics to determine how well a patient would respond to commonly prescribed medications as well as the severity of any side effects.
Published in the journal Cell Genomics, the study found that data from large libraries of sequenced human genomes and other biological data, known as biobanks, can provide new insights into genetic architecture of response to widely prescribed drugs.
Study first author and UCLA Bioinformatics Ph.D. candidate Michal Sadowski said the most common method used to analyze the genetics of drug response is through pharmacogenomic studies in genotyped participants of randomized controlled trials. However, these studies have a small number of participants, are costly and sometimes are not even feasible depending on the drug, Sadowski said.
Genetic data in biobanks provide several benefits. Along with containing sequenced genetic data of large populations, including people both on and off certain medications, these libraries can also be analyzed at a lower cost. While biobank data cannot replace randomized controlled trials, they can unlock new information that can improve future studies and advance the evolving field of using genetics to predict treatment outcomes, Sadowski said.
“We hope that in the future this will enable clinicians and patients to weigh the benefits and risks of a treatment in a more personalized way, and make more informed and timely decisions to embark on the treatment,” Sadowski said. “We expect that the analysis of biobank data will be most useful for widely prescribed drugs.”
The study, supervised by UCLA Neurology, Computational Medicine, and Human Genetics professor Noah Zaitlen and UChicago Genetic Medicine assistant professor Andy Dahl, used genetic data from more than 342,000 people in the UK Biobank. Researchers analyzed how their genetic makeups impacted their response to four of the most commonly prescribed drugs in the world: statins for high cholesterol, metformin for type 2 diabetes, warfarin for blood clots, and methotrexate for autoimmune diseases and cancer.
Sadowski and his colleagues sought to determine how large of a role genetic variation played in the variability in response to these drugs as well as which specific genes were involved.
“If a lot can be explained by genetics, then genetics can be used as a good predictor to how you will respond to the drug,” Sadowski said. “Say you want to take statins because of your cholesterol levels. Your physician can look at your genetics and give you an opinion including on potential side effects. If you have predictors that say you will respond well and there is a low chance that you will have side effects, it's likely a good choice to start the treatment.”
For example, the study identified 156 genes that can potentially drive the variation of statins’ impact on LDL cholesterol levels. In total, about 9% of the variation of drug response was attributed to genetic differences from person to person.
Additionally, the study found that gene-drug interactions can also influence the predictive power of a genetic risk tool known as a polygenic score. Polygenic scores are used to summarize the combined effect of a large number of genetic variants to estimate a person’s risk for developing a certain trait or disease. The models to generate these scores must be trained on genetic data from large populations of people and have important limitations, including being based largely on data from people of European ancestry.
Sadowski’s study found standard polygenic scores’ accuracy was likely to underperform in clinical contexts because it contained data from both statin and non-statin users.
“We were surprised to see that polygenic predictors had such significant differences in performance between people who are on and off drugs,” Sadowski said. “We were also surprised by the magnitude of drug-specific heritability for some outcomes. These collectively suggest that additional genetic associations and components of missing heritability could be revealed through future context-specific studies of complex disease.”
The study has several limitations, with future work needed to improve reliability of inference from observational data from biobanks and to understand the limitations of genetic risk prediction.
A copy of the study can be found at https://www.cell.com/cell-genomics/fulltext/S2666-979X(24)00351-3
