Learn Explainer
DPYD and fluoropyrimidine chemotherapy: four variants, one enzyme and what a DNA file can show
DPYD makes the enzyme that clears most of a dose of fluorouracil or capecitabine. Four variants that weaken it are graded by CPIC and sit behind the EMA's 2020 recommendation to test before treatment. What follows is how that score is built, and which of the four a chip, VCF, gVCF or BAM file can show.
Key takeaways
- DPD breaks down about 80 percent of a fluorouracil dose, and CPIC reports severe toxicity in 10 to 40 percent of fluoropyrimidine-treated patients, with DPD deficiency one cause[1].
- CPIC grades four variants as primary: c.1905+1G>A (*2A) and c.1679T>G (*13) remove function, c.2846A>T and HapB3 reduce it. The phenotype comes from summing two allele values, from 0 to 2[1].
- The EMA's 2020 recommendation is phenotype and/or genotype testing before fluorouracil, capecitabine or tegafur. CPIC's dosing tables cover fluorouracil and capecitabine; tegafur is rated no recommendation[3, 1].
- In gnomAD's Finnish group c.1905+1G>A is on 2.3 percent of alleles and HapB3 on 1.3 percent, the reverse of other Europeans; FinnGen shows the same order[15, 16].
- The four variants explain 20 to 30 percent of early-onset 5-FU toxicity, per CPIC, so "no variant found" is not clearance[1].
- Aimosti reads DPYD fully only from BAM or CRAM, at two of four positions from a VCF or gVCF, and gives a chip file no phenotype unless both copies of c.1905+1G>A show.
Fluorouracil (5-FU) is a chemotherapy drug given by infusion, and capecitabine and tegafur are tablets the body converts to it. About 80 percent of an administered dose is broken down by one enzyme, dihydropyrimidine dehydrogenase (DPD), which the DPYD gene encodes; only 1 to 3 percent turns into the forms that kill cells[1]. When DPD works poorly the drug stays in the blood longer, and the Clinical Pharmacogenetics Implementation Consortium (CPIC) links four DPYD variants to a higher chance of severe, sometimes fatal toxicity[1, 2]. In April 2020 the EMA recommended phenotype or genotype testing before treatment with these drugs[3].
One enzyme clears most of the drug
Fluorouracil, written 5-FU, is given by infusion for solid tumours, among them colorectal and breast cancer and cancers of the aerodigestive tract, mostly inside a combination regimen. Capecitabine is a tablet that the body converts to 5-FU, and tegafur is another prodrug[1]. Together they are the fluoropyrimidines. DPD does the first, rate-limiting step of clearing them[1]. A person with less DPD carries more of a standard dose, for longer.
CPIC puts severe toxicity at 10 to 40 percent of the people treated: low neutrophil counts, vomiting, severe diarrhoea, mouth ulcers, hand-foot syndrome[1]. DPD deficiency is one cause among several. The EMA's product information for capecitabine estimates how common it is.
The four variants CPIC grades as primary
CPIC's current DPYD allele table lists the reference allele and 83 variant alleles, of which 21 remove function, 7 reduce it and 55 behave normally[5]. For 5-FU, CPIC singles out four "due to their population frequency and established impact on enzyme function and toxicity risk"[1]. Two do the most damage, and two do it moderately.
| Variant | rsID | GRCh38 position and letters in a VCF | What it does | CPIC function and value |
|---|---|---|---|---|
| c.1905+1G>A (*2A) | rs3918290 | 1:97450058 C>T | Splice donor change; exon 14 is skipped and the protein does not work | No function, 0 |
| c.1679T>G (*13) | rs55886062 | 1:97515787 A>C | Missense change, p.I560S | No function, 0 |
| c.2846A>T | rs67376798 | 1:97082391 T>A | Missense change, p.D949V | Decreased function, 0.5 |
| c.1129-5923C>G (HapB3) | rs75017182 | 1:97579893 G>C | Deep intronic change that creates a cryptic splice site | Decreased function, 0.5 |
Source: CPIC's 2017 guideline[1] and its database export of 13 June 2026[5]; positions and strand from Ensembl[6]. VCF letters are on the plus strand; variant names are on the gene's coding strand.
HapB3 is a haplotype, a run of linked letters from intron 5 to exon 11, and CPIC treats the intronic c.1129-5923C>G as the likely causal change. The synonymous c.1236G>A (rs56038477) sits on the same haplotype and was in perfect linkage with it in the 1000 Genomes Project, so in Europeans it works as a proxy, and it is the letter the EMA's product information names[1, 4]. One 2025 study found 45 of 46 c.1236G>A heterozygotes carrying the intronic variant too, and one who was homozygous for it, and its authors ask laboratories to read both variants, or at least the intronic one[7].
DPYD lies on the minus strand of chromosome 1[6]. The variant names describe the gene as it is read, and a VCF describes the plus strand, so each letter pair is complemented: c.1905+1G>A is a C>T in the file, c.2846A>T a T>A. CPIC flags this for the two transversions, whose decreased-function letters are G and T on the coding strand[1]. It matters again in section 6.
In heterozygous carriers, DPD activity measured in blood ran 50 percent lower for c.1905+1G>A, 68 percent for c.1679T>G, 30 percent for c.2846A>T and 35 percent for HapB3[1]. A pooled analysis of 7,365 patients from eight studies put the relative risk of severe toxicity at 2.9, 4.4, 3.0 and 1.6 for the same four[1, 2].
From two alleles to an activity score
CPIC does not lean on star alleles for DPYD. Few variants have a star name, and elsewhere a star describes a haplotype of several variants, so the guideline prefers HGVS names or rsIDs[1]. Our star alleles guide covers the convention, and the CYP2D6 guide the other gene where CPIC adds allele values. The DPYD scheme has fewer: 1 for a normal-function allele, 0.5 for decreased function, 0 for none[1, 8].
The gene's activity score is the sum of the two lowest values among the variants found. Poor metabolizers score 0 or 0.5, intermediate metabolizers 1 or 1.5, normal metabolizers 2. Two different reduced-function variants are presumed to sit on different gene copies[1].
CPIC attaches a recommendation to each score. For a normal metabolizer it is the label dose. For an intermediate metabolizer, a 50 percent lower starting dose, titrated against toxicity or drug levels. At 0.5 and 0, avoiding 5-FU and its prodrugs, with, at 0.5 only and if the treating team sees no suitable alternative, a strongly reduced dose and early drug monitoring[1, 5]. These are statements about what the guideline recommends to prescribers. The intermediate line has changed once: the 2017 text gave 25 to 50 percent for a score of 1.5, and in November 2018 CPIC made it 50 percent at both 1 and 1.5, noting that people homozygous for c.2846A>T may warrant a larger cut[1, 9]. The Dutch Pharmacogenetics Working Group (DPWG) also uses 50 percent at 1 and 1.5[10].
A second change is announced and not yet in the tables. CPIC's page carries a July 2026 notice that the guideline now being updated will give c.1129-5923C>G an allele value of 0.75, so a heterozygous carrier scores 1.75 with a cycle 1 start at 75 percent of the intended dose. The tables change only when the guideline is published, expected in fall 2026[9]; this article uses today's values. The notice does not name its evidence. Two Dutch results lie nearby: HapB3-proxy carriers on a 25 percent lower dose had a relative risk of severe toxicity of 1.69 against 1.72 for full-dose carriers in earlier data, and in an exploratory matched analysis their progression-free survival was shorter, hazard ratio 1.43[11, 12].
What the EMA and the product information say about testing
On 30 April 2020 the EMA announced that patients "should be tested for the lack of the enzyme dihydropyrimidine dehydrogenase (DPD)" before cancer treatment with injected or infused fluorouracil, capecitabine or tegafur. The review began in March 2019 at the French medicines agency's request, and the notice says the European Commission would issue the binding decision later[3]. For healthcare professionals the notice words the test this way.
Phenotype and/or genotype testing is therefore recommended before starting treatment with fluoropyrimidines.
The notice adds that treatment is contraindicated in known complete DPD deficiency, that a reduced starting dose should be considered in identified partial deficiency, and that testing is not required before the antifungal flucytosine, which may be urgent. It excludes fluorouracil used on the skin, where little is absorbed[3]. The current Xeloda product information repeats the testing statement and gives uracil thresholds: 16 ng/mL or more and below 150 indicates partial deficiency, 150 or more complete deficiency[4].
| Genotype test | Phenotype test | |
|---|---|---|
| What is measured | DPYD variants, in most tests the four above[1] | Uracil in blood plasma, which DPD breaks down[3, 4] |
| How it is read | Allele values summed into a band[1] | Thresholds of 16 and 150 ng/mL, which the label calls uncertain[4] |
| Evidence cited | 1,103 Dutch patients: 8 percent carried a variant, and severe toxicity ran at 39 percent in carriers on reduced doses against 23 percent in non-carriers[11] | 550 patients: uracil above 16 ng/mL went with an odds ratio of 5.3 for severe toxicity in cycle 1[13] |
| A limit named | Testing selected variants "does not fully rule out DPD defects"[1] | Uracil can read falsely high in moderate kidney impairment[4] |
Source: CPIC 2017[1], the EMA notice[3], the Xeloda product information[4], and the two studies in the cells[11, 13].
A retrospective comparison from Finland's Ostrobothnia hospital district, 80 colorectal cancer patients before routine genotyping and 69 after, reports lower 90-day mortality afterwards and no grade 3 or worse toxicity in carriers started on reduced doses[14]. The groups are small and not randomised.
How common each variant is, and what differs in Finland
For Europeans, CPIC gives carrier frequencies of 4.7 percent for HapB3, 1.6 for c.1905+1G>A and 0.7 for c.2846A>T, and says 7 percent carry at least one of the four[1]. The Xeloda text gives 2.6 to 6.3 percent for HapB3 and around 1 percent for c.1905+1G>A[4]. Neither splits out Finland, so we pulled the four from gnomAD v4.1, which pools more than 800,000 exomes and genomes: its Finnish group, about 32,000 people, and its non-Finnish European group, about 590,000[15]. HapB3 is represented by its proxy c.1236G>A, which has both exome and genome counts.
Read the pairs from the top. In the non-Finnish European group HapB3 supplies about 65 percent of the four variants' alleles and c.1905+1G>A about 14 percent, which fits CPIC's description of HapB3 as the commonest[1]. In the Finnish group the order flips: c.1905+1G>A supplies about 63 percent and HapB3 about 36 (our arithmetic on the counts below).
| Variant | Finnish alleles | Non-Finnish European alleles | Finnish carriers | Other European carriers |
|---|---|---|---|---|
| c.1905+1G>A (*2A) | 1,494 of 64,030 | 5,701 of 1,179,786 | 4.7 | 1.0 |
| HapB3 (c.1236G>A) | 860 of 64,026 | 26,048 of 1,179,710 | 2.7 | 4.4 |
| c.2846A>T | 27 of 64,022 | 7,583 of 1,179,644 | 0.08 | 1.3 |
| c.1679T>G (*13) | 8 of 64,024 | 996 of 1,179,282 | 0.02 | 0.17 |
| All four, added | 7.5 | 6.8 |
Source: gnomAD v4.1 joint exome and genome counts, queried 10 October 2026[15]. Carriers per 100 is our arithmetic: twice the allele frequency, which assumes random pairing and ignores people with two of the variants. It estimates a database group and says nothing about a person.
The carrier column adds to about 7 per 100 in both groups, in line with CPIC's 7 percent; only the mix differs. FinnGen's twelfth release points the same way, with c.1905+1G>A at 2.0 percent, the HapB3 proxy and the intronic HapB3 variant at 1.4 and c.2846A>T at 0.05; c.1679T>G has no record in its browser[16]. We could not check how far FinnGen's participants overlap with gnomAD's Finnish cohorts, so the two corroborate each other without being independent, and neither is a random sample of Finland.
The consequence is about test design. A test reading only HapB3 would see about a third of the variant alleles in the Finnish counts. One reading c.1905+1G>A and c.2846A>T but not HapB3, as Aimosti's variant-file module does, would see about 64 percent of them in the Finnish group and about 33 percent in the non-Finnish European group (our arithmetic, on alleles rather than people).
What a chip can and cannot say about DPYD
All four variants are single-letter changes, which is what a genotyping chip measures. Two obstacles remain: whether the array carries the position, which varies by array and version, and strand. A chip reports a pair of letters, and for a change between A and T or between C and G the pair does not show which strand it was read from. Two of the four are in that class, c.2846A>T and HapB3's intronic c.1129-5923C>G[1, 5].
The two homozygous genotypes change meaning between strands; a heterozygote reads the same on both. Aimosti's chip module never reads an A/T or C/G position as a variant, whatever the file says, so rs67376798 is never called from a chip. The other two variants, c.1905+1G>A (C>T) and c.1679T>G (A>C), are not ambiguous, and our chip module reads the first and not the second. It reads neither HapB3 letter. A chip file therefore has one resolvable position of the four.
What Aimosti's report shows for each file
| File | What the DPYD card shows | What it cannot show |
|---|---|---|
| Chip export (23andMe, AncestryDNA, MyHeritage) | No phenotype, with one exception. The card names what was read at c.1905+1G>A and says rs67376798 cannot be oriented. If both copies of c.1905+1G>A show, it calls a poor metabolizer, because the open position cannot change that | c.2846A>T, HapB3, c.1679T>G |
| Plain VCF | A phenotype from CPIC's table, from c.1905+1G>A and c.2846A>T. With neither found, the card reads as the reference version and says a variant-only file cannot prove the positions were covered | HapB3, c.1679T>G |
| gVCF | The same two positions. Reference blocks show whether they were called, so a reference result carries a shorter caveat, and an explicit no-call leaves the gene unresolved | HapB3, c.1679T>G |
| BAM or CRAM (Deep Read) | A diplotype from PyPGx across the whole gene, looked up in a table built from CPIC's 84 alleles. A diplotype outside the table is reported as indeterminate | Stretches the reads did not cover; the card reports the callable share |
Source: How the report is built today, pinned by tests/test_article_dpyd_fluoropyrimidines.py. PyPGx 0.26.0, the pinned version, names 83 DPYD alleles including HapB3 and c.1679T>G[17]; the lookup table is CPIC's[5].
A variant-file result is a lookup on two variants, so a "normal metabolizer" line there means neither was found and says nothing about HapB3. The cards restate CPIC's text for fluorouracil and capecitabine, attributed to CPIC, and say nothing about tegafur, for which CPIC has no recommendation[5]. No module measures uracil or DPD activity.
Our DPYD gene page and capecitabine and fluorouracil page restate CPIC's guidance in the report's form, and the guide to VCF, gVCF, BAM and CRAM files covers what each file keeps. A DNA-file result is not the clinical test.
What Aimosti would (and wouldn't) show you
A chip export gets no DPYD phenotype, because rs67376798 cannot be oriented from a chip; the one exception is a chip that shows both copies of c.1905+1G>A, which we call a poor metabolizer. A plain VCF and a gVCF are read at two positions, c.1905+1G>A and c.2846A>T, and looked up in CPIC's table. A BAM or CRAM runs Deep Read, which calls the whole gene with PyPGx, HapB3 and c.1679T>G included. Each card restates CPIC's wording for fluorouracil and capecitabine and says nothing about tegafur.
What we won't claim
We won't call a DPYD result clearance for a fluoropyrimidine, give a dose, say whether anyone should take a drug, or read a position the file cannot orient. Whether to test before treatment, and what to do with a result, is a decision between a patient and their oncology team, and the clinical test is the result they work from.
Bottom line. A chip file shows almost none of the four DPYD variants, a variant file two, and aligned reads all of them, and even all four leave most early severe toxicity unexplained. A DNA-file result is background to the clinical test.
Questions people ask
Is DPYD in my 23andMe or AncestryDNA raw data?
Some positions may be, but c.2846A>T and HapB3's intronic letter are A/T or C/G changes, where a chip's letter pair does not show the strand. Aimosti gives no DPYD phenotype from a chip file except where both copies of c.1905+1G>A show, and does not read HapB3 or c.1679T>G from a chip.
What is the difference between a DPYD genotype test and a DPD phenotype test?
A genotype test reads DPYD variants, usually the four CPIC grades, and sums them into an activity score. A phenotype test measures uracil in blood plasma; the Xeloda product information gives 16 and 150 ng/mL as thresholds for partial and complete deficiency[4]. The EMA recommends "phenotype and/or genotype testing"[3].
Does a normal DPYD result rule out severe toxicity?
CPIC says no. Patients without a decreased or no function variant may still have severe toxicity from other genetic, environmental or other factors, and the four variants explain 20 to 30 percent of early-onset 5-FU toxicities[1].
Why does the EMA recommendation name tegafur when CPIC gives no dosing for it?
The EMA's testing recommendation covers tegafur[3]. CPIC's dosing tables apply only to fluorouracil and capecitabine, because tegafur is given with uracil or with gimeracil and oteracil, which inhibit DPD, and evidence on DPYD variants in that setting is very limited[1]. The DPWG, which does cover tegafur, recommends avoiding it at scores of 0, 1 and 1.5[10].
References
- Amstutz U, Henricks LM, Offer SM, et al. Clinical Pharmacogenetics Implementation Consortium (CPIC) Guideline for Dihydropyrimidine Dehydrogenase Genotype and Fluoropyrimidine Dosing: 2017 Update. Clinical Pharmacology and Therapeutics, 2018. doi:10.1002/cpt.911 PMID 29152729. Read in full from the PDF CPIC hosts. Table 1 gives the phenotype bands, Table 2 the dosing, and the text the four primary variants, carrier frequencies and caveats quoted here.
- Meulendijks D, Henricks LM, Sonke GS, et al. Clinical relevance of DPYD variants c.1679T>G, c.1236G>A/HapB3, and c.1601G>A as predictors of severe fluoropyrimidine-associated toxicity: a systematic review and meta-analysis of individual patient data. The Lancet Oncology, 2015. doi:10.1016/S1470-2045(15)00286-7 PMID 26603945. 7,365 patients from eight studies; adjusted relative risk 4.40 for c.1679T>G and 1.59 for c.1236G>A/HapB3.
- EMA recommendations on DPD testing prior to treatment with fluorouracil, capecitabine, tegafur and flucytosine. European Medicines Agency, 2020. Press release of 30 April 2020, reference EMA/229267/2020, read in full on 10 October 2026.
- Xeloda: EPAR product information (summary of product characteristics), section 4.4. European Medicines Agency, 2026. Read on 10 October 2026. Section 4.4, DPD deficiency: complete deficiency 0.01 to 0.5 percent and partial 3 to 9 percent of Caucasians; heterozygous frequencies around 1 percent for c.1905+1G>A, 1.1 for c.2846A>T, 2.6 to 6.3 for c.1236G>A/HapB3, 0.07 to 0.1 for c.1679T>G; uracil 16 ng/mL and 150 ng/mL thresholds.
- CPIC DPYD alleles, gene-drug pairs and recommendations. Clinical Pharmacogenetics Implementation Consortium, 2026. Read from CPIC's database export of 13 June 2026 (api.cpicpgx.org/v1): 84 DPYD alleles, of which 21 no function, 7 decreased function and 56 normal function; fluorouracil and capecitabine at level A, tegafur at level C.
- Ensembl gene ENSG00000188641 (DPYD). Ensembl, 2026. Read through the Ensembl REST service on 10 October 2026: GRCh38, chromosome 1, 97,077,743 to 97,995,000, strand -1; rs3918290 splice donor variant, rs67376798 missense variant, rs75017182 intron variant.
- Gil-Rodriguez A, Recarey-Rama S, Fernandez Montes A, et al. A Lack of Complete Linkage Disequilibrium Between c.1236G>A and c.1129-5923C>G HapB3 Variants of DPYD: A Call to Revise European Pharmacogenetic Guidelines. International Journal of Molecular Sciences, 2025. doi:10.3390/ijms26178136 PMID 40943060. 45 of 46 c.1236G>A heterozygotes kept complete linkage; one was homozygous for c.1129-5923C>G.
- Henricks LM, Lunenburg CA, Meulendijks D, et al. Translating DPYD genotype into DPD phenotype: using the DPYD gene activity score. Pharmacogenomics, 2015. doi:10.2217/pgs.15.70 PMID 26265346.
- CPIC guideline for DPYD and fluoropyrimidines: online updates since publication. ClinPGx, 2026. Read on 10 October 2026 through the ClinPGx API. Entries dated November 2018 (50 percent reduction for scores 1 and 1.5), January 2024 (HapB3 and c.1129-5923C>G split in the allele tables) and July 2026 (announced allele value 0.75 for c.1129-5923C>G, not yet in the tables).
- Lunenburg CATC, van der Wouden CH, Nijenhuis M, et al. Dutch Pharmacogenetics Working Group (DPWG) guideline for the gene-drug interaction of DPYD and fluoropyrimidines. European Journal of Human Genetics, 2020. doi:10.1038/s41431-019-0540-0 PMID 31745289. Gene activity score 0 to 2; 50 percent of the standard dose at scores 1 and 1.5; tegafur to be avoided at 0, 1 and 1.5.
- Henricks LM, Lunenburg CATC, de Man FM, et al. DPYD genotype-guided dose individualisation of fluoropyrimidine therapy in patients with cancer: a prospective safety analysis. The Lancet Oncology, 2018. doi:10.1016/S1470-2045(18)30686-7 PMID 30348537. 1,103 evaluable patients in 17 Dutch hospitals; 85 (8 percent) heterozygous carriers; severe toxicity 39 percent in carriers and 23 percent in wild-type patients; relative risk 1.69 for c.1236G>A carriers on a 25 percent lower dose.
- Knikman JE, Wilting TA, Lopez-Yurda M, et al. Survival of Patients With Cancer With DPYD Variant Alleles and Dose-Individualized Fluoropyrimidine Therapy: A Matched-Pair Analysis. Journal of Clinical Oncology, 2023. doi:10.1200/JCO.22.02780 PMID 37639651. Exploratory analysis; 61 c.1236G>A carriers matched to three controls each; progression-free survival hazard ratio 1.43 (95 percent CI 1.10 to 1.86).
- Meulendijks D, Henricks LM, Jacobs BAW, et al. Pretreatment serum uracil concentration as a predictor of severe and fatal fluoropyrimidine-associated toxicity. British Journal of Cancer, 2017. doi:10.1038/bjc.2017.94 PMID 28427087. 550 patients; uracil above 16 ng/mL, odds ratio 5.3 for severe toxicity in the first cycle.
- Saarenheimo J, Willfor H, Wahid N, Jekunen A, Andersen H. Impact of Upfront DPYD Genotyping on Fluoropyrimidine Adjuvant Therapy in Colorectal Cancer: A Real-World Data. Clinical Colorectal Cancer, 2025. doi:10.1016/j.clcc.2025.02.001 PMID 40000255. Abstract read; the full text is not open access. Retrospective, Wellbeing Services County of Ostrobothnia: 80 patients in 2016 to 2018 and 69 in 2020 to 2022.
- gnomAD v4.1: allele counts for four DPYD variants, Finnish and non-Finnish European groups. Genome Aggregation Database, Broad Institute, 2026. Queried on 10 October 2026 through the browser's GraphQL API, dataset gnomad_r4, joint exome and genome counts. Variants 1-97450058-C-T, 1-97515787-A-C, 1-97082391-T-A, 1-97573863-C-T and 1-97579893-G-C (GRCh38). Counts are stored in data/articles/dpyd-fluoropyrimidines/allele-frequencies.json.
- FinnGen R12 public browser: DPYD variant pages. FinnGen, 2026. Read on 10 October 2026 through the browser's API: 1:97450058:C:T (rs3918290) allele frequency 0.020118; 1:97573863:C:T (rs56038477) 0.013985; 1:97579893:G:C (rs75017182) 0.013969; 1:97082391:T:A (rs67376798) 0.000479; 1,045,686 alleles counted. rs55886062 has no record.
- Lee SB. PyPGx: DPYD allele table, version 0.26.0. GitHub, 2026. Version 0.26.0 is the one Aimosti's worker pins. Listing its DPYD alleles on 10 October 2026 returned 83, among them c.1905+1G>A (*2A), c.1679T>G (*13), c.2846A>T and c.1129-5923C>G, c.1236G>A (HapB3).
Last reviewed . Every number on this page links to the source it comes from; if one of them has moved, tell us.
Free · in your browser
Find out what your own file can read.
The free check opens your DNA file on your device and shows which format you have and what a report could read from it. The file is never uploaded.