Loading field map
For decades, clinicians have observed that the same dose of a drug can be life-saving for one patient and toxic for another. The search for the biological roots of this variability has driven a half-century of genetic inquiry, evolving from the study of single inherited traits to the analysis of entire genomes and their biological networks. Pharmacogenomics, as the field is now known, sits at the intersection of pharmacology and genomics, asking how genetic variation shapes drug response—and whether that knowledge can make therapy safer and more effective.
The earliest attempts to link genetics to drug response focused on rare, dramatic reactions that ran in families. In the 1950s, researchers noticed that some patients given the muscle relaxant succinylcholine remained paralyzed for hours, a trait later traced to a deficiency in the enzyme butyrylcholinesterase. Around the same time, the antimalarial drug primaquine was found to cause hemolytic anemia in individuals with glucose-6-phosphate dehydrogenase (G6PD) deficiency, an X-linked condition. These discoveries gave rise to Pharmacogenetics, a framework that treated drug response as a simple Mendelian trait: a single gene variant explains the outcome. The method was hypothesis-driven—researchers selected a candidate gene based on known pharmacology (e.g., a drug-metabolizing enzyme) and tested whether its variants correlated with drug levels or adverse effects. This approach produced clinically useful predictions for a handful of drugs, such as thiopurine methyltransferase (TPMT) testing for azathioprine toxicity and CYP2D6 genotyping for codeine efficacy. Yet Pharmacogenetics remained confined to a narrow set of pathways; it could not explain the many cases where drug response varied without a clear family history or a strong single-gene effect. The deterministic view—one gene, one drug effect—proved too simple for most medications.
The completion of the Human Genome Project in 2003, together with the development of high-throughput genotyping arrays, transformed the search for genetic predictors. Instead of guessing which gene mattered, researchers could scan the entire genome for statistical associations between common variants (single-nucleotide polymorphisms, or SNPs) and drug response. This framework, Pharmacogenomics, replaced the candidate-gene method with hypothesis-free genome-wide association studies (GWAS). The shift was not merely a change in technique; it involved a new way of thinking about genetic architecture. Pharmacogenomics recognized that most drug responses are polygenic—influenced by many variants, each of small effect—and that even well-characterized pharmacogenetic loci like CYP2C9 and VKORC1 for warfarin dosing account for only a fraction of the observed variability. By 2009, GWAS had identified dozens of novel loci for drugs ranging from statins (SLCO1B1 and myopathy risk) to antiplatelet agents (CYP2C19 and clopidogrel efficacy). Pharmacogenomics did not replace Pharmacogenetics in the clinic; rather, it coexisted with it, adding a discovery engine that fed new candidate variants back into clinical testing. The older candidate-gene tests remained in use for high-penetrance variants, while genome-wide data began to inform polygenic models.
As GWAS data accumulated, researchers faced a puzzle: how to use the many small-effect variants that individually fail to reach statistical significance? The solution was to combine them into a single quantitative measure—a Polygenic Risk Score (PRS). First applied to common diseases in 2009, PRS was quickly adapted for drug response. The method sums the number of risk or effect alleles, weighted by their estimated effect sizes from a discovery GWAS, to produce a score that stratifies patients across the population. For drug response, PRS can predict, for example, the likelihood of achieving a target international normalized ratio (INR) on warfarin, or the risk of major adverse cardiovascular events on clopidogrel, by aggregating many variants beyond the canonical pharmacogenetic loci. PRS extended the Pharmacogenomics framework by providing a way to capture polygenic architecture in a single predictor, but it also narrowed the focus from mechanistic understanding to statistical prediction. A PRS does not reveal which biological pathways are involved; it is a black-box composite that may generalize poorly across ancestries. This limitation fueled the development of a more mechanistic alternative.
Beginning around 2010, a growing number of researchers argued that the focus on statistical association, whether single-gene or polygenic, missed the deeper biology. Drug response does not arise from isolated genetic variants but from the interactions of many genes in regulatory and signaling networks. Systems Pharmacogenomics emerged as a framework that places genetic variation in the context of biological pathways, protein–protein interactions, and expression quantitative trait loci (eQTL). Instead of testing each SNP individually, systems approaches use network-propagation algorithms to prioritize variants that converge on the same functional modules, or integrate transcriptomic and proteomic data to explain how a variant alters pathway activity. For example, a systems pharmacogenomic analysis of statin response might identify not only SLCO1B1 but also a network of lipid-metabolism genes whose combined variation better predicts myopathy risk than any single locus. Systems Pharmacogenomics also revives the mechanistic ambition of earlier candidate-gene work, but with a richer tool kit: it assumes that the relevant unit of analysis is the pathway, not the gene. This framework does not replace PRS; rather, it competes for clinical adoption by offering interpretability at the cost of statistical simplicity.
Today, all four frameworks remain active, each carving out a distinct niche. Pharmacogenetics continues to guide clinical decisions for high-penetrance variants—TPMT, CYP2D6, DPYD, and others—and is embedded in drug labels. Pharmacogenomics provides the discovery engine for new variants, though its clinical use is largely indirect, feeding into PRS or systems models. Polygenic Risk Scores have gained traction for common, complex drug responses where many small effects sum to a meaningful prediction, but they face challenges in cross-ancestry transportability and regulatory acceptance. Systems Pharmacogenomics, the most recent addition, aims to deliver mechanistic explanations that can point to new drug targets or combination therapies, yet it remains largely a research tool due to the complexity of network models and the need for multi-omic data.
The four frameworks agree on a central premise: genetic variation influences drug response, and incorporating it into prescribing decisions can improve outcomes. They disagree, however, about what kind of evidence is sufficient for clinical action. Pharmacogenetics and Pharmacogenomics emphasize reproducible statistical associations; PRS accepts predictive accuracy even without mechanistic understanding; Systems Pharmacogenomics insists that a predictor should also explain biology. This tension between prediction and explanation is unlikely to be resolved soon. In practice, many clinical pharmacogenomics programs now use a hybrid approach: they test a panel of well-validated single-gene variants (Pharmacogenetics), calculate a PRS for drugs where such scores have been validated, and consult network models for drug development rather than routine care. The history of the field is thus not a linear succession of better frameworks, but an expanding toolbox whose pieces are deployed depending on the drug, the patient, and the question being asked.