Learn Explainer
CYP2D6 and your medicines: star alleles, gene copies and what a DNA file can show
CYP2D6 is the liver enzyme that turns codeine into morphine and tamoxifen into endoxifen, and it clears a long list of antidepressants. CPIC uses it in its prescribing recommendations for nineteen drugs. Reading the gene is harder than reading almost any other on a prescribing list, because a person can carry no copy of it, one, two, three or more, and most DNA files cannot count.
Key takeaways
- CPIC uses CYP2D6 in its recommendations for nineteen drugs, among them codeine, tramadol, tamoxifen, ondansetron, atomoxetine, paroxetine and several tricyclic antidepressants[3].
- A CYP2D6 phenotype comes from an activity score: each allele carries a value of 0, 0.25, 0.5 or 1, a duplicated allele counts once per copy, and the total falls into one of four bands, poor, intermediate, normal or ultrarapid[11, 12].
- Gene deletions, duplications and hybrids with the neighbouring CYP2D7 gene decide a large part of the score. A genotyping chip and a variant file record letters at positions, not the number of copies, so they cannot see these changes.
- By our estimate from CPIC's allele frequencies, about 6.5 percent of people of European ancestry are poor metabolizers and 2.3 percent ultrarapid; in CPIC's Oceanian group the ultrarapid share is 17.8 percent[9].
- Aimosti reads CYP2D6 only from a BAM or CRAM, where read depth lets the caller count copies. Chips, plain VCFs and gVCFs get no CYP2D6 result, and a diplotype with an allele outside our activity table is reported as indeterminate.
CYP2D6 is one of the cytochrome P450 enzymes the liver uses to process medicines. For some drugs it does the activating: codeine has little effect until CYP2D6 converts part of it to morphine, and tramadol and tamoxifen depend on it in the same way[1, 2]. For others it does the clearing, so slower enzyme means more drug left in the blood. The Clinical Pharmacogenetics Implementation Consortium (CPIC), which writes gene-based prescribing guidelines, lists nineteen drugs for which it uses CYP2D6 in its recommendations and fifty-nine more where it has scored the evidence but makes no recommendation[3].
Nineteen guideline drugs run through one enzyme
Whether a slow CYP2D6 is a problem depends on which job the enzyme does for a given drug. When the drug is a prodrug, an inactive form the body has to convert, a slow enzyme means too little of the active drug and a fast one means too much. Codeine is the standard example. CPIC's opioid guideline gives the implication for an ultrarapid metabolizer as "increased formation of morphine leading to higher risk of toxicity", and for a poor metabolizer as "greatly reduced morphine formation leading to diminished analgesia"[1, 4]. Tramadol behaves the same way through its active metabolite, and tamoxifen, the breast cancer drug, needs CYP2D6 to make endoxifen[1, 2].
When CYP2D6 clears a drug instead, the directions reverse. A poor metabolizer keeps more of it in the blood, which CPIC links to a higher chance of side effects for paroxetine and the tricyclic antidepressants, and an ultrarapid metabolizer clears it faster, which for ondansetron, the anti-sickness drug, CPIC associates with a weaker response[4, 5, 6, 7]. The table lists the guidelines CPIC has written with CYP2D6 in them.
| Guideline | Drugs where CYP2D6 informs the recommendation | What CYP2D6 does to the drug | Our report restates it |
|---|---|---|---|
| Opioids (2021) | Codeine, tramadol, hydrocodone | Activates it | Codeine and tramadol |
| Tamoxifen (2018) | Tamoxifen | Activates it | Yes |
| Ondansetron and tropisetron (2017) | Ondansetron, tropisetron | Clears it | No |
| Tricyclic antidepressants (2016 update) | Amitriptyline, nortriptyline, clomipramine, desipramine, doxepin, imipramine, trimipramine | Clears it | No |
| Serotonin reuptake inhibitors (2023) | Paroxetine, fluvoxamine, vortioxetine, venlafaxine | Clears it; for venlafaxine, makes its active metabolite | Yes |
| Atomoxetine (2019) | Atomoxetine | Clears it | No |
| Beta-blockers (2024) | Metoprolol | Clears it | No |
Source: CPIC's gene-drug pairs and guideline citations, from its database export of 13 June 2026[3]; guideline papers[1, 2, 5, 6, 7, 8]. The last column is what the Deep Read card restates today.
Some of these are the antidepressants people ask about most, and our separate guide to gene tests for antidepressants goes through what the trials of such testing found. Here the point is narrower: for paroxetine, vortioxetine and the tricyclics, CPIC's recommendation changes with the CYP2D6 phenotype, while for citalopram, escitalopram and sertraline it is CYP2C19 that CPIC uses[3, 5]. CPIC also lists drugs such as pitolisant, eliglustat and tetrabenazine with a CYP2D6 evidence level but no CPIC recommendation[3].
Star alleles and the activity score
Versions of the gene are named as star alleles. CYP2D6*1 is the reference version with normal function, *4 is the commonest version with none in people of European ancestry[9], *5 means the whole gene is missing from that chromosome, and an x2 after a name means the allele is present twice in a row[10]. PharmVar, the consortium that assigns the names, defines each allele by the set of changes it carries[10]. CPIC then gives each allele a function and an activity value; its current table holds 207 CYP2D6 allele records, including duplicated and hybrid forms[11].
Terms this article relies on
- Star allele
- A named version of a pharmacogene, such as
CYP2D6*4, defined by the variants it carries on one chromosome. - Diplotype
- The two alleles a person carries, one per chromosome, written
*1/*4. A duplication is written on the side it sits on:*1x2/*4. - Activity value
- CPIC's number for how much working enzyme one allele contributes: 1 for normal function, 0.5 or 0.25 for decreased function, 0 for none[11].
- Activity score
- The sum of the activity values in a diplotype, counting each copy of a duplicated allele[12].
- Copy number
- How many copies of the gene a person has across both chromosomes. Most people have two; a deletion lowers the count and a duplication raises it.
In 2019 CPIC and the Dutch Pharmacogenetics Working Group agreed one way to turn a CYP2D6 genotype into a phenotype, so that the same genotype would get the same phenotype from both[12]. Under that consensus a score of 0 is a poor metabolizer, 0.25 to 1 intermediate, 1.25 to 2.25 normal, and anything above 2.25 ultrarapid, and the *10 allele, common in East Asia, counts 0.25[11, 12]. CPIC has since moved *9 and *41 from 0.5 to 0.25 as well[11].
*1 counting twice, and the bands are CPIC's[11, 12].The arithmetic is simple. What it needs as input is the hard part: the right alleles, and the right number of them.
Our Deep Read card scores the alleles below. It holds the common ones; anything else the reads show is reported without a phenotype, for reasons the next sections go into.
| Allele | Function | Activity value |
|---|---|---|
CYP2D6*1 | Normal function | 1 |
CYP2D6*2 | Normal function | 1 |
CYP2D6*3 | No function | 0 |
CYP2D6*4 | No function | 0 |
CYP2D6*5 | No function | 0 |
CYP2D6*6 | No function | 0 |
CYP2D6*9 | Decreased function | 0.25 |
CYP2D6*10 | Decreased function | 0.25 |
CYP2D6*17 | Decreased function | 0.5 |
CYP2D6*29 | Decreased function | 0.5 |
CYP2D6*41 | Decreased function | 0.25 |
Source: Read at load time from the Deep Read content file, which a test pins to CPIC's allele table[11]. A called allele outside this list makes the card indeterminate.
Gene copies, the part a variant list cannot hold
CYP2D6 sits on chromosome 22 next to two close relatives, CYP2D7 and CYP2D8, which no longer make a working enzyme but share much of its sequence[10]. That neighbourhood rearranges. A whole copy of CYP2D6 can be lost, which is the *5 allele, or repeated, which gives *1x2, *2x2 and the other duplications. Pieces of CYP2D7 can also end up in a CYP2D6 copy, making hybrid genes such as *13 and *36; *36 is often found in tandem with a *10, written *36+*10[10, 11].
A genotyping chip measures single letters at fixed positions, and a variant file lists the letters where a genome differs from the reference. Neither records how many copies of a stretch of DNA were present. Someone with one copy of the gene and a deletion on the other chromosome has only one version of each position to read, so a variant caller sees what looks like two identical copies. Someone with a duplication gives the same picture at every position the copies share. PharmVar's review of the gene makes the point about the deletion: a sample carrying *5 on both chromosomes may be called as two ordinary copies when the sequencing data were processed without a structural-variant caller[10].
Reads solve the counting problem by comparing depth over CYP2D6 with depth over a gene that is reliably present in two copies. The caller our Deep Read runs, PyPGx, does this and then classifies the copy-number profile with a model trained on real and simulated samples, because the likeness of CYP2D6 and CYP2D7 makes some reads land on the wrong gene[13]. Our pipeline uses VDR as the control gene.
How often each phenotype turns up
Gaedigk and colleagues used CPIC's allele frequency tables, then drawn from more than 60,000 people in 173 reports, to predict phenotypes across world populations. They found poor metabolizers in 0.4 to 5.4 percent of each population and ultrarapid metabolizers in 1 to 21 percent, using the activity-score cut-offs of the time[14]. CPIC's tables have grown since then, and the 2019 consensus moved the cut-offs.
6.5%
of people in CPIC's European group predicted to be poor metabolizers, by our estimate[9]
17.8%
predicted ultrarapid in CPIC's Oceanian group, by our estimate[9]
0.4 to 5.4%
poor metabolizers across world populations in the 2017 analysis[14]
We repeated the calculation on CPIC's current frequencies. For each population group, every pair of alleles is weighted by how common the two are, its activity values summed and placed in a band[9]. Pairs that include an allele CPIC has not scored are left unassigned, between 2.5 and 10 percent depending on the group. The result is an estimate for a population, never for a person.
| CPIC population group | Gene deleted (*5) | Working gene duplicated | Predicted poor | Predicted ultrarapid |
|---|---|---|---|---|
| African American/Afro-Caribbean | 5.4% | 2.8% | 2.3% | 4.1% |
| American | 1.6% | 3.3% | 2.0% | 5.2% |
| Central/South Asian | 4.2% | 1.3% | 2.4% | 1.5% |
| East Asian | 4.8% | 0.9% | 0.8% | 0.9% |
| European | 2.9% | 1.9% | 6.5% | 2.3% |
| Latino | 2.9% | 2.9% | 3.1% | 4.1% |
| Near Eastern | 1.8% | 6.5% | 2.2% | 7.4% |
| Oceanian | 3.5% | 11.9% | 0.3% | 17.8% |
| Sub-Saharan African | 6.2% | 3.0% | 2.0% | 4.5% |
Source: Allele frequencies from CPIC's database export of 13 June 2026[9]. The deletion and duplication columns are summed CPIC allele frequencies; the predicted columns are our estimate, assuming alleles pair at random.
The groups are broad. CPIC's "European" group pools many studies, and Finland is not split out in these tables, so the European row is the nearest stand-in for a Finnish reader rather than a Finnish figure[9]. The estimate also inherits CPIC's assumptions: an allele scored 0.25 is treated as exactly that, whatever the drug.
What our report does with each kind of file
| File | CYP2D6 result | Why |
|---|---|---|
| Chip export (23andMe, AncestryDNA, MyHeritage) | None; the medication section says the gene needs aligned reads | A chip reads fixed letters and cannot count copies |
| Plain VCF | None, with the same note | Lists letters where the genome differs; no copy count, and silence elsewhere |
| gVCF | None, with the same note | Proves which positions were read, but still holds letters, not copies |
| BAM or CRAM (Deep Read) | Diplotype, copy-number call, activity score, phenotype and CPIC guidance for codeine, tramadol and tamoxifen | Read depth against a control gene gives the copy count |
Source: How the report is built today. The gVCF tier calls seventeen other pharmacogenes with PyPGx and leaves CYP2D6 out by design.
The gVCF row deserves a word, because a gVCF is the best variant file there is: it records where the sequencing looked as well as what it found, and our report uses that to call seventeen pharmacogenes from it. The same software can produce a CYP2D6 diplotype from a gVCF too. We tested that against the reads on one genome: the variant-only call dropped a duplication without a warning, so the activity score came out one whole working copy short. So the gene stays on the reads path, where the copy count is measured. Our guide to VCF, gVCF, BAM and CRAM files goes through what each file keeps.
What Deep Read does with CYP2D6
- Checks the depth. A genome with a mean depth below 5 reads is refused before any calling, because copy number cannot be estimated from that little data. GRCh37 and GRCh38 files are both read.
- Counts the copies. PyPGx compares the depth over CYP2D6 with the depth over VDR, a control gene, and classifies the copy-number profile[13]. The result appears on the card as the structural call, such as a whole-gene duplication.
- Names the diplotype. The same run reads the variants in the gene and assembles the two alleles, duplications and deletions included, into a diplotype.
- Checks the gene was readable. If less than 30 percent of the gene is covered by at least 10 reads, the card says so and gives no phenotype.
- Scores it, or says it cannot. The activity values of the named alleles are added and banded. If any allele is outside the card's table, the hybrids
*13,*36and*68among them, the card reports the gene as indeterminate instead of giving a phenotype. If PyPGx returns no call at all, the card says the gene was not called.
The last step is a real limit. CPIC gives *36+*10 an activity value of 0.25 and *13 and *68 a value of 0[11], and the reads can show them, but our card does not score them yet and says so on the card. An indeterminate card means our table could not score what the reads showed; it says nothing about how well the enzyme works.
What Aimosti would (and wouldn't) show you
Deep Read calls CYP2D6 from a BAM or CRAM file with PyPGx, which estimates the number of gene copies from read depth against a control gene and returns a diplotype such as *1x2/*4. The card adds up CPIC's activity values for the alleles it names, gives the phenotype CPIC's cut-offs assign to that score, and restates CPIC's guidance for it on codeine, tramadol, tamoxifen, paroxetine, fluvoxamine, vortioxetine and venlafaxine. A diplotype containing an allele outside the card's activity table, which includes the hybrid alleles, is reported as indeterminate. A chip export, a plain VCF and a gVCF do not produce a CYP2D6 result at all: the report's medication section leaves the gene out and says it needs the aligned reads.
What we won't claim
We won't infer a CYP2D6 result from a file that cannot count gene copies, assign a phenotype to an allele CPIC has not scored in our table, or turn a guideline into an instruction. What a phenotype means for a prescription is a decision for the prescriber who knows the rest of the picture.
Bottom line. CYP2D6 matters to more CPIC guidelines than almost any other gene, and its result rests on a count of gene copies. A file that keeps the reads can settle that count; a chip or variant file cannot, and a CYP2D6 result built from one is a guess about the part that decides the answer.
Questions people ask
Is CYP2D6 in my 23andMe or AncestryDNA raw data?
A few positions inside the gene may be, which is enough to see some single-letter alleles such as *4. The deletions, duplications and CYP2D7 hybrids that move the activity score are not, because a chip does not count copies. Our report does not give a CYP2D6 result from a chip file.
Can a gVCF from my sequencing provider give a CYP2D6 result?
Software can produce one, but it assumes two copies of the gene. When we compared it with the reads on one genome, it missed a duplication. Aimosti calls CYP2D6 from the BAM or CRAM, where read depth gives the copy count; the gVCF still gives seventeen other pharmacogenes.
What does a CYP2D6 activity score of 1 mean?
Under the CPIC and DPWG consensus it is an intermediate metabolizer, typically one normal allele and one that does not work, such as *1/*4[12]. Scores from 1.25 to 2.25 are normal, 0 is poor, and anything above 2.25 is ultrarapid.
Does CYP2D6 affect antidepressants?
For some. CPIC uses CYP2D6 in its recommendations for paroxetine, fluvoxamine, vortioxetine and venlafaxine, and for tricyclics such as amitriptyline and nortriptyline; for citalopram, escitalopram and sertraline it uses CYP2C19 instead[5, 6]. Our Deep Read card names the CYP2D6 phenotype and restates CPIC's guidance for paroxetine, fluvoxamine, vortioxetine and venlafaxine, along with codeine, tramadol and tamoxifen; it carries none for the tricyclics.
Why do some reports call CYP2D6 from a chip anyway?
Because most people do have two copies, so a call that assumes it is often right. The file cannot show when it is wrong. In CPIC's European frequencies the *5 deletion alone is on about 2.9 percent of chromosomes[9], and every carrier of it looks like a person with two copies on a chip.
References
- Crews KR, Monte AA, Huddart R, et al. Clinical Pharmacogenetics Implementation Consortium Guideline for CYP2D6, OPRM1, and COMT Genotypes and Select Opioid Therapy. Clinical Pharmacology and Therapeutics, 2021. PMID 33387367.
- Goetz MP, Sangkuhl K, Guchelaar HJ, et al. Clinical Pharmacogenetics Implementation Consortium (CPIC) Guideline for CYP2D6 and Tamoxifen Therapy. Clinical Pharmacology and Therapeutics, 2018. PMID 29385237.
- CPIC gene-drug pairs and levels for CYP2D6. Clinical Pharmacogenetics Implementation Consortium, 2026. Read from CPIC's database export of 13 June 2026 (api.cpicpgx.org/v1, pair table): 19 CYP2D6 pairs used for recommendations, 59 more with a CPIC level and no recommendation.
- CPIC recommendations for CYP2D6 by phenotype. Clinical Pharmacogenetics Implementation Consortium, 2026. Read from CPIC's database export of 13 June 2026 (recommendation table).
- Bousman CA, Stevenson JM, Ramsey LB, et al. Clinical Pharmacogenetics Implementation Consortium (CPIC) Guideline for CYP2D6, CYP2C19, CYP2B6, SLC6A4, and HTR2A Genotypes and Serotonin Reuptake Inhibitor Antidepressants. Clinical Pharmacology and Therapeutics, 2023. doi:10.1002/cpt.2903 PMID 37032427.
- Hicks JK, Sangkuhl K, Swen JJ, et al. Clinical Pharmacogenetics Implementation Consortium (CPIC) Guideline for CYP2D6 and CYP2C19 Genotypes and Dosing of Tricyclic Antidepressants: 2016 Update. Clinical Pharmacology and Therapeutics, 2017. PMID 27997040.
- Bell GC, Caudle KE, Whirl-Carrillo M, et al. Clinical Pharmacogenetics Implementation Consortium (CPIC) guideline for CYP2D6 genotype and use of ondansetron and tropisetron. Clinical Pharmacology and Therapeutics, 2017. PMID 28002639.
- Brown JT, Bishop JR, Sangkuhl K, et al. Clinical Pharmacogenetics Implementation Consortium Guideline for Cytochrome P450 (CYP)2D6 Genotype and Atomoxetine Therapy. Clinical Pharmacology and Therapeutics, 2019. PMID 30801677.
- CPIC allele frequencies for CYP2D6 by population group. Clinical Pharmacogenetics Implementation Consortium, 2026. Read from CPIC's database export of 13 June 2026 (allele table, frequency field). The predicted shares in the table are our arithmetic on these frequencies.
- Nofziger C, Turner AJ, Sangkuhl K, et al. PharmVar GeneFocus: CYP2D6. Clinical Pharmacology and Therapeutics, 2020. doi:10.1002/cpt.1643
- CPIC allele functionality and activity values for CYP2D6. Clinical Pharmacogenetics Implementation Consortium, 2026. Read from CPIC's database export of 13 June 2026 (allele and diplotype tables): 207 CYP2D6 allele records; *9 and *41 at 0.25.
- Caudle KE, Sangkuhl K, Whirl-Carrillo M, et al. Standardizing CYP2D6 Genotype to Phenotype Translation: Consensus Recommendations from the Clinical Pharmacogenetics Implementation Consortium and Dutch Pharmacogenetics Working Group. Clinical and Translational Science, 2020. doi:10.1111/cts.12692
- Lee SB. PyPGx: a Python package for predicting PGx genotypes and phenotypes from sequencing data. GitHub, 2026. README, section Structural variation detection, read 9 October 2026.
- Gaedigk A, Sangkuhl K, Whirl-Carrillo M, Klein T, Leeder JS. Prediction of CYP2D6 phenotype from genotype across world populations. Genetics in Medicine, 2017. doi:10.1038/gim.2016.80 PMID 27388693.
Last reviewed . Every number on this page links to the source it comes from; if one of them has moved, tell us.