Celiac Disease Genes in Your Raw DNA Data: HLA-DQ2.5 (rs2187668), DQ8 (rs7454108) and What They Can't Tell You

Your 23andMe or AncestryDNA raw file contains the two variants that tag the main celiac disease haplotypes: rs2187668 for HLA-DQ2.5 and rs7454108 for HLA-DQ8. Search the file for both, read the pair off the table below, and you know whether you carry them. Most carriers never develop celiac disease, so the useful answer is a negative one; a positive one is a question for a doctor, not a diet change.

Why most people with the celiac genes never get celiac disease#

Celiac disease is an immune reaction to gluten that damages the lining of the small intestine. It affects about 1% of people worldwide, many of them undiagnosed (NIDDK, checked 14 October 2026).

It almost never occurs in people who lack one of two groups of HLA variants, DQ2 and DQ8, which shape the immune molecules that present gluten fragments to T cells. The variants are common. NIDDK says "about 30 percent of people have DQ2 or DQ8" and "only about 3 percent of people with DQ2 or DQ8 develop celiac disease" (NIDDK, checked 14 October 2026); a 2020 review puts the carrier share at 30% to 40% of the general population in North America and Europe, with around 3% of carriers developing the disease (Poddighe et al., World Journal of Gastroenterology 2020). The arithmetic matters: if a third of people carry the variants and 1% have the disease, about 3 carriers in 100 are affected and 97 are not. The genes are close to necessary but far from sufficient; gluten is the trigger, and the rest is not well understood (Aboulaghras et al., International Journal of Molecular Sciences 2023).

DQ2 comes in several forms. DQ2.5 (the DQA1*05:01 and DQB1*02:01 pair) is found in about 90% of patients and DQ8 (DQB1*03:02) in about 5% (Aboulaghras et al. 2023); almost 95% of patients carry at least one of the two (Monsuur et al., PLoS ONE 2008). The rest mostly carry DQ2.2 or half of the DQ2 molecule, which matters for the gap described below.

The two tag SNPs: rs2187668 and rs7454108#

A consumer chip does not read the HLA genes themselves. It reads single positions nearby that are inherited together with a haplotype, called tag SNPs. Monsuur and colleagues showed that six such SNPs predict the DQ2.5, DQ2.2, DQ7 and DQ8 types with a sensitivity above 0.991 and a specificity above 0.996 against classical HLA typing (Monsuur et al. 2008). Two of the six are the ones 23andMe reads in its Celiac Disease report (23andMe for Healthcare Professionals, October 2017, checked 14 October 2026):

• rs2187668 tags DQ2.5. It sits in HLA-DQA1 on chromosome 6, position 32,605,884 on build GRCh37, the build consumer files use, with alleles C and T (Ensembl GRCh37, checked 14 October 2026). Monsuur's table gives T as the tag allele. SNPedia describes the SNP in minus orientation and calls the same allele A (SNPedia rs2187668), as does the GWAS Catalog, but 23andMe "reports all genotype data based on the plus strand of GRCh37" (SNPedia orientation page), so in your file the DQ2.5 allele is T.
• rs7454108 tags DQ8. It lies near HLA-DQB1 at position 32,681,483 on GRCh37, alleles T and C (Ensembl GRCh37). It is a plus-orientation SNP, so C tags DQB1*03:02 in your file, on SNPedia and in the catalog alike (SNPedia rs7454108).

The table uses the letters as a 23andMe file prints them, in alphabetical order, so expect CT rather than TC. The frequencies are the HapMap CEU percentages on the two SNPedia pages, a small European sample. SNPedia's chip tags list both SNPs on 23andMe's v3 to v5 chips and on Ancestry's v2 chips, so a file from either company should contain them.

CC at the first and TT at the second means neither haplotype, the result most worth having, for the reason given under the limitations.

The DQ2.2 gap, and what else two SNPs miss#

Two SNPs cover DQ2.5 and DQ8. They do not detect DQ2.2 (the DQA1*02:01 and DQB1*02:02 pair), the lower-risk form of DQ2 that accounts for most patients who have neither main haplotype. In a European cohort of 1,008 celiac patients, 61 (6%) lacked both DQ2.5 and DQ8; 57 of those carried half of the DQ2 molecule, and only 4 patients in the whole cohort carried none of it (Karell et al., Human Immunology 2003).

Calling DQ2.2 from a raw file takes three SNPs read together, rs2395182, rs7775228 and rs4713586, the panel Monsuur used; SNPedia's genoset gs221 encodes the combination (SNPedia gs221). SNPedia lists the third on 23andMe's v5 chip only (not v3 or v4) and on Ancestry's v2 chips, and it is a minus-orientation SNP whose letters must be flipped first, so we give no DQ2.2 table here. The practical consequence: CC at rs2187668 plus TT at rs7454108 means "no DQ2.5 and no DQ8", not "no celiac genes". 23andMe's report has the same two-variant scope; its footnote says the test "is not intended to diagnose any disease" and that the report "does not cover non-celiac disease gluten sensitivity" (23andMe for Healthcare Professionals).

How to find them in your file without uploading it#

1. Open the file in a plain text editor. Unzip the download (our download guide covers each provider) and open the .txt (23andMe, AncestryDNA) or .csv (MyHeritage) in TextEdit, Notepad or a code editor, not Excel, which can reformat values.
2. Search for rs2187668. A 23andMe line looks like rs2187668 6 32605884 CT; an AncestryDNA line splits it into two columns, C and T. Note the letters, then search for rs7454108.
3. Read both off the table above. If a line shows -- (23andMe) or 0 0 (Ancestry), the chip did not call that position and the file cannot answer.

Our guide to reading a raw DNA data file explains the columns, the GRCh37 build, strand orientation and no-calls.

What the GWAS evidence says, and how our report handles it#

The GWAS Catalog, the database behind our reports, listed 16 associations for rs2187668 (GWAS Catalog rs2187668, checked 14 October 2026). Two are celiac disease: an odds ratio of 7.04 in the first celiac GWAS (van Heel et al., Nature Genetics 2007) and 6.23 in its 2010 successor (Dubois et al., Nature Genetics 2010). Most of the rest are immune conditions that travel with the same haplotype, such as lupus, autoimmune hepatitis and sarcoidosis. rs7454108 has 4, among them a vasculitis row (odds ratio 2.73), the one that reaches our catalog (GWAS Catalog rs7454108). An odds ratio of 7 describes the odds across carriers against non-carriers, not a 7-fold risk for you; the 3% figure above is what it looks like in absolute terms.

Our health catalog (release 2026-06-01) keeps the rows that pass the methodology filter: seven for rs2187668 (both celiac rows, lupus, sarcoidosis, autoimmune hepatitis, collagenous colitis and cutaneous lupus) and one for rs7454108 (vasculitis). The celiac condition itself rests on 39 rows across 33 variants, and the nutrition catalog's Gluten Sensitivity entry draws on 161 rows whose mapped trait is celiac disease.

Two caveats about how those rows reach your PDF. First, our generator compares the catalog's risk allele letter with the letters in your file as printed, with no strand flipping. The catalog spells the rs2187668 risk allele A, the minus-strand letter, while your file prints C or T, so the two celiac rows, the lupus row and the sarcoidosis row cannot match a 23andMe or AncestryDNA file; two more rows report no allele and never match; only the cutaneous lupus row (reported as T) matches a DQ2.5 carrier. The rs7454108-C row matches a CT or CC file as expected. In practice, the celiac verdict in our report comes from the other celiac variants, not from your DQ2.5 status; read that status from the table above instead. Second, the condition's text is filed under the Congenital section of the complete health report, while collagenous colitis appears under Gastrointestinal and cutaneous lupus, autoimmune hepatitis and vasculitis under Autoimmune and Inflammatory Diseases. Each row shows the rsID, risk allele, your genotype, odds ratio and PubMed link (see the sample report); the verdict multiplies odds ratios without weighting one copy against two, without correcting for correlated variants and without a baseline risk, so it is an association, not a risk score. As with APOE, the row is what tells you something.

What the result cannot tell you#

• A tag is not the gene. Monsuur's panel was developed in Dutch cohorts and checked in British, Spanish and Italian samples; linkage between a tag SNP and a haplotype can differ elsewhere, which is why 23andMe says its report is most relevant for people of European descent.
• The negative is the useful half. HLA testing for celiac disease "has a poor positive predictive value but a high negative predictive value" (Aboulaghras et al. 2023); NIDDK puts it as "if you do not have these gene variants, you are very unlikely to have celiac disease" and "having DQ2 or DQ8 alone does not mean you have celiac disease" (NIDDK diagnosis page, checked 14 October 2026). With the DQ2.2 gap, a consumer-file negative is a little weaker than a laboratory one.
• Symptoms are a job for serology, not a raw file. The first-line test is the tissue transglutaminase IgA antibody (tTG-IgA), positive in about 93% of patients on a gluten-containing diet and negative in about 96% of people without the disease; a small-intestine biopsy confirms it (Celiac Disease Foundation, checked 14 October 2026). The antibody test only works while you eat gluten, and NIDDK notes that doctors "don't recommend starting a gluten-free diet before diagnostic testing". If you have symptoms, see a doctor or gastroenterologist, with the two lines from your file as context. Each haplotype came from a parent, so a positive result is information about your siblings and children too.

FAQ#

Does 23andMe test for celiac disease?

Yes, in two ways. The Celiac Disease Genetic Health Risk report, one of the ten reports the FDA authorised on 6 April 2017 (23andMe press release, checked 14 October 2026), reads rs2187668 and rs7454108 and is sold, at the time of writing, only with the Health service in eight countries. Separately, every customer's raw file contains both rsIDs, so an ancestry-only kit still gives you the two letters.

How many genes are involved in celiac disease?

HLA-DQ is by far the strongest, which is why two tag SNPs are informative at all. Beyond it, dense genotyping in 2011 brought the number of known risk loci, including HLA, to 40, each of the others with a small effect (Trynka et al., Nature Genetics 2011). Our health catalog holds 39 celiac rows across 33 variants, each linked to its study.

Can you have celiac disease without the gene?

Rarely. In 1,008 European patients, 6% lacked both DQ2.5 and DQ8, but nearly all of those carried half of the DQ2 molecule, and only 4 patients carried none of it (Karell et al. 2003). The two consumer SNPs do not see that half-DQ2 or DQ2.2 group, so "negative in my file" is not quite "no celiac genes"; a laboratory HLA test is the stricter version.

Should I stop eating gluten if I carry DQ2 or DQ8?

Not on the basis of the gene alone. About 97% of carriers never develop celiac disease, and going gluten-free before testing makes the antibody test unreliable, which can leave a real case undiagnosed. If you have symptoms or a family history, ask a doctor for a tTG-IgA test while you still eat gluten, and bring the two lines from your file to that appointment.