Most preclinical models ask whether manipulating a target changes disease-like biology in a laboratory system. Human genetics asks a more consequential question: when nature perturbs that target across real people, does disease risk or a disease-relevant trait change? That is why genetic support has become one of the most valuable forms of target-validation evidence in modern drug discovery.
Analyses by Matthew Nelson and colleagues, and later by Eric King and colleagues, found that drug mechanisms with human genetic support were associated with a higher probability of clinical success. The exact uplift depends on definitions, datasets, and development stage, so “twice as likely” should not be treated as a universal constant. The direction is nevertheless important: targets anchored in human biology tend to carry better priors than targets supported only by convenient models.
The PCSK9 story shows why genetics is powerful
The PCSK9 pathway is an unusually clean demonstration. Gain-of-function variants were associated with high LDL cholesterol, while loss-of-function variants were associated with lower LDL and protection from coronary heart disease. The human allelic series did more than identify an association. It indicated direction: reducing PCSK9 activity should lower LDL. It also supplied a partial safety experiment because some people lived with substantially reduced PCSK9 function.
This evidence helped make PCSK9 inhibition a compelling therapeutic strategy. But the lesson is not “find a gene near a disease signal and make an inhibitor.” The lesson is to build a causal chain from variant to molecular consequence, target activity, intermediate phenotype, disease outcome, and safety.
A genetic association is not automatically a target mechanism
Genome-wide association studies often identify regions containing multiple correlated variants and several plausible genes. The nearest gene is not necessarily the causal gene. The associated variant may alter an enhancer that acts at a distance, affect expression only in a specific cell state, or tag another unmeasured causal variant. Fine mapping, colocalization with molecular QTLs, functional genomics, perturbation experiments, and disease-relevant cellular models are needed to move from locus to mechanism.
Target validation therefore has at least three separate questions:
- Causal gene: Which gene or regulatory element mediates the association?
- Direction: Should therapeutic intervention increase, decrease, or reshape its activity?
- Context: In which tissue, cell type, developmental stage, and patient subgroup does the effect operate?
Lifelong genetics is not the same intervention as a drug
A person carrying a variant has experienced that perturbation from conception. A drug may be started at age sixty after disease is established. Developmental compensation may reduce or amplify the phenotype of a genetic variant. A partial lifelong reduction in protein abundance is not always equivalent to acute catalytic inhibition. A variant that changes one domain may not mimic degradation of the entire protein. Pharmacology also introduces dose, schedule, tissue distribution, metabolites, and off-target effects that genetics does not reproduce.
Human genetics can validate the direction of a biological hypothesis without validating every possible way of pharmacologically implementing it.
This is where many superficially “genetically validated” programmes become less certain. The target may be right while the modality is wrong. The mechanism may be beneficial in one tissue and harmful in another. The relevant biology may require intervention before irreversible pathology develops.
Pleiotropy can create false confidence
Variants often influence more than one biological pathway. Mendelian randomization is powerful when its assumptions are credible, but horizontal pleiotropy can make an apparent causal relationship reflect another pathway carried by the same genetic instrument. Linkage disequilibrium can also make two nearby signals appear to be one. Robust analysis therefore uses multiple variants, sensitivity analyses, colocalization, biological annotation, and triangulation with independent evidence rather than relying on a single headline association.
Population diversity is part of target validity
Genetic architecture, allele frequency, linkage patterns, environmental exposures, and disease prevalence vary across populations. Evidence discovered in one ancestry may transfer biologically while the predictive marker does not transfer statistically. Underrepresentation can also hide protective alleles or safety-relevant biology. A target programme should record where its genetic evidence comes from and how confidently it generalizes.
Use genetics to sharpen experiments, not replace them
The strongest workflow treats human genetics as a high-value prior that changes what is tested next. An allelic series can suggest the desired direction and magnitude of modulation. Tissue-specific expression and single-cell data can identify the relevant cellular context. Perturbation studies can test whether the proposed causal gene reproduces the human phenotype. Biomarker strategy can be designed around the genetically implicated pathway. Safety analysis can search for phenotypes associated with stronger or opposite perturbation.
Human genetics is often described as nature's randomized trial. The metaphor is useful because inheritance precedes disease and can reduce confounding. But nature does not randomize a clean, selective drug at a chosen dose for twelve weeks. Genetics raises the quality of the question. Translational pharmacology still has to build the answer.
