From star-allele diplotype to activity score, PBPK exposure curves, and CPIC-guided dosing across PM/IM/NM/UM phenotypes
CYP2D6 is arguably the single most consequential pharmacogene in clinical use, metabolizing roughly 20–25% of all commonly prescribed drugs — including opioids, antidepressants, antipsychotics, and tamoxifen. Yet it is also one of the most structurally complex loci in the human genome: it sits adjacent to two nonfunctional pseudogenes (CYP2D7, CYP2D8) with which it readily recombines, producing gene deletions, duplications, and hybrid alleles that defeat simple SNP genotyping. Accurate calling requires a combination of targeted variant panels and copy-number-aware sequencing.
CYP2D6 sits in a genomically unstable region on chromosome 22, flanked by two closely related, nonfunctional pseudogenes (CYP2D7 and CYP2D8) that share >90% sequence identity. Non-allelic homologous recombination between these paralogs is common and produces:
• Whole-gene deletions (*5 allele) — no protein product • Whole-gene duplications/multiplications (*1xN, *2xN, up to 13 copies reported) — increased protein dose • Gene-conversion hybrids (*36, *68, *83) — chimeric CYP2D6/CYP2D7 sequences with altered or absent function • Tandem arrangements — a duplication allele carrying one functional and one nonfunctional copy in cis, which must be distinguished from two separate functional alleles
Because of this, single-SNP genotyping (e.g., a 2–3 marker panel) is inadequate for CYP2D6 in a way it is not for most other pharmacogenes. Clinical-grade panels test 20–25+ core variants simultaneously (TaqMan OpenArray, Luminex xTAG CYP2D6 Kit, or Agena iPLEX), and reference laboratories increasingly use long-read technologies.
02 — Long-read sequencing and copy-number confirmation: • PacBio HiFi or Oxford Nanopore sequencing spans the full ~4.3 kb gene plus flanking pseudogene homology regions in single reads, directly resolving phase and structural variants that short-read NGS cannot • Droplet digital PCR (ddPCR) with gene-specific and reference-assay probes gives absolute copy-number quantification (0, 1, 2, 3+ copies) with coefficient of variation <5% • Multiplex ligation-dependent probe amplification (MLPA) is a lower-cost alternative for exon-level deletion/duplication screening • Sample QC gates: DNA concentration >10 ng/µL, A260/280 1.7–2.0, call rate >99% before diplotype assembly proceeds
Raw genotype calls are meaningless to a clinician without translation into the star-allele nomenclature curated by PharmVar (Pharmacogene Variation Consortium), the community standard that superseded the original CYP allele nomenclature website in 2018. Each star allele (e.g., *1, *4, *10, *17) represents a specific haplotype — a defined combination of variants inherited together on one chromosome — with an assigned functional status. Correct diplotype assembly requires resolving which variants are in cis (same chromosome) versus trans (opposite chromosomes), a step that ordinary genotyping cannot do without additional phasing information.
A patient heterozygous for both the 1846G>A splice variant (defining *4) and the −1584C>G promoter variant (defining *41) could in principle carry *4/*41 (one nonfunctional, one decreased-function chromosome) or a rarer configuration where both variants sit on the same chromosome, leaving the other chromosome fully wild-type. These two scenarios predict different phenotypes and different drug doses, but produce identical unphased genotype output.
Phasing methods used clinically: • Long-read single-molecule sequencing (PacBio HiFi, ONT) — direct physical linkage of variants on one DNA molecule, the current gold standard • Family/trio genotyping — parental data constrains possible phase • Allele-specific long-range PCR — amplifies each chromosome copy separately before variant detection • Statistical population-based phasing (as a fallback) — lower confidence, flagged for confirmatory testing
02 — Diplotype-to-nomenclature translation pipeline: • Variant calls are matched against the PharmVar core allele definition table (core + suballeles, e.g., *2.001–*2.079) • CNV/structural results are integrated: a duplication is denoted with a multiplication suffix, e.g., *1x2, *2x3 • Automated calling software (Stargazer, Aldy, Cyrius) cross-validates against the ambiguous-call table maintained by CPIC, flagging diplotypes that require manual review • Final report format follows the standard convention: [allele1]/[allele2], e.g., *4/*41, *1/*1xN, *5/*10
The Activity Score (AS) system, adopted by CPIC and the Dutch Pharmacogenetics Working Group (DPWG), assigns each CYP2D6 allele a numeric function value and sums the two allele values in a diplotype into a single continuous score. This transforms a categorical genotype into a quantitative proxy for enzyme activity, allowing the four classical phenotype categories — poor, intermediate, normal, and ultra-rapid metabolizer — to be defined by simple, guideline-standardized AS thresholds.
CPIC (2024 update) defines four phenotype categories from the summed activity score:
• Poor Metabolizer (PM): AS = 0 — both alleles nonfunctional (e.g., *4/*4, *4/*5, *3/*4). Frequency ~5–10% in European ancestry populations, lower in most other groups. • Intermediate Metabolizer (IM): 0 < AS ≤ 1.0 — one nonfunctional plus one normal or decreased-function allele, or two decreased-function alleles (e.g., *4/*41, *10/*10). Frequency ~10–15%. • Normal Metabolizer (NM): 1.0 < AS ≤ 2.25 — the reference/majority phenotype (e.g., *1/*1, *1/*2, *1/*41). Frequency ~65–75%. • Ultra-Rapid Metabolizer (UM): AS > 2.25 — typically a functional allele duplication or multiplication (e.g., *1x2/*1, *2x3/*4). Frequency 1–2% in most populations but 10–28% in parts of Northeast Africa and the Arabian Peninsula, driven by regionally high CYP2D6 duplication rates.
The AS system replaced a purely categorical "PM/IM/EM/UM" call because it captures dose-response gradation: a *1/*41 diplotype (AS=1.5, low-normal) behaves measurably differently from a *1/*1 diplotype (AS=2.0), even though both were historically lumped as "extensive/normal metabolizer."
The table below lists commonly encountered alleles, their molecular basis, and CPIC function assignment used to compute the activity score.
| Product | Indication | Trial Design | Key Result |
|---|---|---|---|
| *1, *2, *35 | Wild-type / normal-function SNPs | No catalytically disruptive change; normal protein expression and activity | Function value = 1.0 |
| *10, *17 | Missense (P34S; T107I/R296C) | Reduced catalytic efficiency / protein stability | Function value = 0.25–0.5 |
| *41 | Intronic splice variant (−1584C>G, linked) | Reduced mRNA splicing efficiency, ~50% normal expression | Function value = 0.5 |
| *3, *4, *5, *6 | Frameshift / splice-defect / whole-gene deletion | No functional protein produced | Function value = 0 |
| *1xN, *2xN | Whole-gene duplication/multiplication | Multiple functional gene copies transcribed | Function value = 1.0 × N copies |
Knowing a patient's activity score is only useful if it can be translated into a predicted drug or metabolite concentration over time. Physiologically based pharmacokinetic (PBPK) platforms such as Simcyp and PK-Sim scale liver CYP2D6 abundance and intrinsic clearance according to phenotype, then simulate absorption, distribution, hepatic first-pass extraction, and elimination through a virtual population, generating concentration-time curves that quantify how dramatically exposure to both parent drug and active metabolite differs across PM, IM, NM, and UM individuals.
PBPK models represent CYP2D6-mediated metabolism using enzyme kinetic parameters — Km (substrate concentration at half-maximal velocity) and Vmax (maximal reaction velocity) — measured in vitro using recombinant CYP2D6 or human liver microsomes. Vmax is scaled per individual by predicted hepatic CYP2D6 abundance, which is itself scaled by genotype-derived activity score:
CLint(phenotype) = CLint(reference NM) × (AS_patient / AS_reference)
This intrinsic clearance feeds into a well-stirred liver model accounting for hepatic blood flow, fraction unbound, and first-pass extraction, producing a full concentration-time profile after oral dosing. For a bioactivated prodrug like codeine:
• PM (AS=0): CYP2D6 pathway near-absent; codeine is shunted to inactive glucuronidation; morphine AUC is ~10–20% of normal metabolizers → inadequate analgesia • UM (AS>2.25): excess enzyme activity converts codeine to morphine rapidly and in excess quantity → morphine Cmax and AUC substantially elevated → risk of respiratory depression, historically fatal in breastfed infants of UM mothers
For an inactivated substrate like metoprolol (parent drug is the active moiety): • PM individuals show 4–5× higher metoprolol AUC and prolonged half-life → excessive β-blockade, bradycardia, hypotension risk at standard doses • UM individuals clear metoprolol unusually fast → subtherapeutic exposure at standard doses
02 — Model validation: • Simulated exposure curves are validated against published clinical PK studies stratified by genotype (e.g., Kirchheiner et al. dose-adjustment reviews) • Sensitivity analysis varies Km/Vmax within literature-reported ranges to generate confidence bounds around the predicted AUC ratio • Virtual bioequivalence trials (n=50–100 simulated subjects per phenotype arm) predict the probability that a given fixed dose will fall outside the therapeutic window for each phenotype group
The entire genotyping-to-PBPK pipeline exists to answer one clinical question: what should this specific patient be prescribed? The Clinical Pharmacogenetics Implementation Consortium (CPIC) publishes peer-reviewed, evidence-graded guidelines translating CYP2D6 phenotype directly into prescribing actions for more than a dozen drug classes, and the FDA maintains a Table of Pharmacogenomic Biomarkers in Drug Labeling that cross-references over 20 CYP2D6-affected drugs.
CPIC guidelines (updated periodically, codeine/opioid guideline most recently revised 2021) specify concrete actions:
• Codeine & tramadol (bioactivated opioids): PM — avoid, use non-CYP2D6-dependent analgesic (e.g., morphine, non-opioid); UM — avoid, particularly contraindicated in nursing mothers and children due to fatal respiratory depression risk in the infant/child; IM — use with caution, consider alternative; NM — standard dosing appropriate • Tamoxifen (breast cancer endocrine therapy): PM — consider alternative endocrine therapy (e.g., aromatase inhibitor) due to significantly reduced endoxifen formation and associated higher recurrence risk in some studies; IM — consider alternative or increased monitoring • Tricyclic antidepressants (amitriptyline, nortriptyline): PM — reduce starting dose by ~50%, use therapeutic drug monitoring; UM — consider alternative drug or substantially increased dose with monitoring, standard doses often subtherapeutic • Metoprolol / other CYP2D6-cleared beta-blockers: PM — consider 50% dose reduction or alternative agent not dependent on CYP2D6; monitor heart rate/blood pressure closely • Atomoxetine (ADHD): PM — reduce dose, slower titration; exposure can be 5–10× higher than NM at standard dosing
All recommendations are graded by strength of evidence (CPIC uses "strong," "moderate," "optional") and are designed for pre-emptive panel-based testing — ordering the genotype once, before any of these drugs is prescribed, so the result is already in the chart when needed.
Despite strong evidence, CYP2D6-guided prescribing faces practical barriers:
• Turnaround time: reactive (post-prescription) genotyping often takes 3–5 days, too slow for acute pain or psychiatric crisis management — driving adoption of pre-emptive panel testing embedded in the EHR with clinical decision support alerts • Phenoconversion: concurrent use of strong CYP2D6 inhibitors (e.g., paroxetine, fluoxetine, bupropion, quinidine) can pharmacologically convert a genotypic NM or UM patient into a phenotypic PM — genotype alone is necessary but not always sufficient for a real-time activity prediction • Polygenic and environmental factors: liver disease, pregnancy, and drug-drug interactions all modulate actual enzyme activity independent of germline genotype • Reimbursement and infrastructure: not all health systems have validated LDTs or reimbursed CPT codes for CYP2D6 CNV-aware panels
Despite these barriers, CYP2D6 remains the most mature and most frequently cited pharmacogene in clinical implementation programs (e.g., Vanderbilt PREDICT, St. Jude PG4KDS, Mayo Clinic RIGHT protocol), and pre-emptive multi-gene panels increasingly return CYP2D6 phenotype into the EHR before the first relevant prescription is ever written.
The FDA issued a boxed warning in 2013, later strengthened in 2017, restricting codeine (and in 2017 tramadol) use in children under 12 and in breastfeeding mothers, directly citing CYP2D6 ultra-rapid metabolizer status after multiple documented pediatric deaths from morphine overdose in breastfed infants of UM mothers taking codeine. This remains the single most-cited real-world case demonstrating that a pharmacogenomic variant can be directly life-threatening rather than merely a matter of drug efficacy.