Pharmacogenomics: Why the Same Drug Works Differently in Different People

Two people receive the same drug, at the same dose, for the same condition. One tolerates it and benefits. The other develops muscle pain within weeks and quietly abandons the prescription — or takes an antiplatelet drug faithfully after a coronary stent and has a thrombotic event anyway. The explanation usually offered is vague: sensitivity, bad luck, poor adherence. Sometimes that is right. But a meaningful fraction of the variation is inherited, predictable and measurable before the first tablet is swallowed. The discipline that measures it is pharmacogenomics.

The premise is unglamorous. A drug is absorbed, carried into tissues, chemically modified by enzymes, and cleared. Nearly every step in that chain is performed by a protein encoded by a gene that varies between people. Inherit a less active enzyme and the drug accumulates; inherit an extra functional copy and it may be cleared before it can act. Where the drug is a prodrug — inert until an enzyme converts it into the active compound — the logic inverts: the fast metaboliser is the one exposed to risk, and the slow metaboliser gets no effect at all.

What the variation actually changes

Two protein families do most of the clinically relevant work. The cytochrome P450 enzymes — CYP2C19, CYP2D6, CYP2C9 and their relatives — perform the conversions that activate or inactivate a large share of prescribed drugs. The membrane transporters, such as OATP1B1 (encoded by SLCO1B1), decide whether a drug is pulled into the organ where it acts or left circulating, where it may do harm.

Inherited variants in these genes are described in star-allele notation — CYP2C19*2, CYP2D6*4 — and the combination of the two copies you carry is translated into a functional category:

• Poor metaboliser — two non-functional or severely reduced copies; almost no enzyme activity.
• Intermediate metaboliser — one impaired copy; partial activity.
• Normal metaboliser — the reference state, and the implicit assumption behind every standard dose.
• Rapid or ultra-rapid metaboliser — increased-function variants or extra gene copies; unusually fast conversion.

These categories are convenient approximations of what is really a continuum, and genotype predicts enzyme behaviour far more reliably for some genes than for others.

Where the evidence is strongest

Pharmacogenomics is often discussed as though every gene–drug pair were equally supported. They are not. A small number are backed by consistent data, a clear mechanism and formal prescribing guidance; the rest range from promising to speculative. The serious list is short.

Clopidogrel is the clearest cardiovascular example. It is a prodrug, and CYP2C19 performs the step that produces the active metabolite. Loss-of-function alleles are not rare — a substantial minority of people of European ancestry carry at least one, and they are considerably more common in East Asian populations. Poor metabolisers achieve materially less platelet inhibition, and after coronary stenting this translates into more thrombotic events. The drug label warns of this, and alternative antiplatelet agents that do not depend on CYP2C19 exist.

In oncology and immunology the stakes are higher still. Patients with deficient TPMT or NUDT15 activity cannot clear thiopurines normally and can develop life-threatening marrow suppression at conventional doses. Patients with reduced DPYD activity cannot break down fluorouracil or capecitabine, with severe and occasionally fatal toxicity as the result — which is why European regulators now advise assessing DPD status before fluoropyrimidine chemotherapy. Testing for HLA-B*57:01 before prescribing abacavir, which can provoke a severe hypersensitivity reaction in carriers, has been routine for years.

Warfarin is the instructive counter-example. CYP2C9 and VKORC1 variants genuinely explain a large share of the variation in dose requirement, yet trials of genotype-guided initiation produced mixed results against careful clinical dosing. Strong biology does not automatically produce a strong clinical trial.

Statins, muscle symptoms and SLCO1B1

Muscle symptoms are the most common reason patients stop the most evidence-backed preventive drug we have. Most of those symptoms are not in fact caused by the statin — blinded rechallenge studies have shown this repeatedly — but some are, and there is a mechanism.

Statins act inside the liver, and OATP1B1 is the transporter that carries them there. A reduced-function variant in SLCO1B1 means less drug enters the hepatocyte and more remains in the circulation, including in skeletal muscle. A genome-wide analysis of a large simvastatin outcome trial identified this variant as the dominant genetic signal for statin myopathy, and it remains one of the most robust findings in the field. The effect is strongest for simvastatin and smaller for other statins — a matter of exposure, not idiosyncrasy.

The practical value lies in the choice of agent and starting dose, not in whether to treat at all. A reduced-function result argues for a statin less dependent on that transporter — not for abandoning lipid-lowering therapy, a distinction that gets lost surprisingly often.

Codeine, tramadol and the CYP2D6 problem

Codeine is, pharmacologically speaking, a delivery system for morphine. It has little analgesic activity of its own; CYP2D6 converts it. Tramadol behaves similarly. This creates a hazard at both ends of the distribution.

Ultra-rapid metabolisers — people carrying extra functional copies of CYP2D6 — convert codeine to morphine faster than expected and can reach dangerous opioid concentrations at ordinary doses. Deaths in children after tonsillectomy, and in a breastfed infant whose mother was an ultra-rapid metaboliser, led regulators to contraindicate codeine in children and during breastfeeding. Poor metabolisers have the opposite problem: almost no morphine is produced, they get no meaningful pain relief, and they are sometimes mislabelled as drug-seeking when they ask for something stronger.

Pharmacogenomics is not fringe science

This is worth stating directly, because the field sits close to the direct-to-consumer genetics market and is often assumed to share its standards. It does not. The Clinical Pharmacogenetics Implementation Consortium publishes peer-reviewed, freely available prescribing guidelines for specific gene–drug pairs, graded by evidence strength; the Dutch Pharmacogenetics Working Group does the same in Europe. Regulators in the United States and Europe include pharmacogenomic information in the labelling of a large number of drugs, and in a handful of cases — abacavir, thiopurines, fluoropyrimidines — testing before prescribing is now expected practice rather than an enhancement.

What a panel does not tell you

A pharmacogenomic panel is a pharmacokinetic instrument. It tells you, with varying confidence, what your body is likely to do to a drug. It says very little about what the drug will do to you.

• It rarely predicts efficacy. SLCO1B1 status tells you about myopathy risk, not about how far your ApoB will fall. Response usually depends on receptor biology and disease mechanism that no metaboliser category captures.
• Phenoconversion is real. A genetically normal CYP2D6 metaboliser taking a strong inhibitor — certain antidepressants, some antifungals — behaves as a poor metaboliser while that drug is on board. Genotype is fixed; phenotype is not.
• Evidence quality varies enormously. The gap between DPYD and the average psychiatric gene–drug claim is not a matter of degree — yet panels often report both with the same visual confidence.
• Ancestry matters. Variant frequencies differ between populations, and some panels test allele sets developed mainly in European cohorts.

And a result never overrides clinical judgement. It is one input among the indication, the comorbidities, the other prescriptions, and the patient in front of you.

Why testing once is the sensible model

Almost everything in medicine has to be repeated. Lipids drift, glucose answers to last week's eating, blood pressure varies by the hour. Your CYP2C19 genotype does not change — including at three in the morning, forty years from now, when you are admitted with an acute coronary syndrome.

A pharmacogenomic result is one of the very few laboratory values you need to obtain only once. Its usefulness depends entirely on whether it is in the record on the day someone reaches for the prescription pad.

That is the argument for testing pre-emptively rather than reactively. A CYP2C19 genotype ordered after a stent thrombosis is too late to have helped; an SLCO1B1 result produced after a patient has concluded that statins are intolerable rarely brings them back. The value is realised at the moment of prescribing, so the result has to already exist — in the record, and interpretable by whoever writes the prescription years later.

This is also why the field belongs in preventive medicine and not only in hospital practice. Prevention is largely long-term pharmacotherapy in people who are currently well — lipid lowering above all — and tolerance decides whether someone stays on treatment across the two decades over which the benefit accrues. A drug discontinued in month three delivers no risk reduction at all.

What to ask your physician

• Ask whether any drug you take long-term has a gene–drug pair with formal guidance. The list is short and specific, and clopidogrel and statins are on it.
• Ask what the panel actually covers. A useful one reports the well-established genes with graded evidence; a weaker one reports dozens of associations with equal apparent authority.
• Ask for the result to be documented where it will be found. A report held only by the laboratory does nothing at the moment it is needed.
• Mention new prescriptions at review. Interacting drugs can convert a normal metaboliser into a functional poor one, and the genotype report will not warn you.

Pharmacogenomics does not personalise medicine in the sweeping sense the phrase usually implies. It resolves a narrow, well-defined set of questions about dose and drug choice — and it resolves them permanently. For a small number of drugs, that is the difference between a therapy that works and one that harms.