Can a test tell you whether gluten got through?

Fifty-two adults with coeliac disease were given a cookie. Some cookies held gluten, some held none, and nobody — not the participants, not the researchers — knew which was which. Then everyone tried to work out who had been exposed.

The question sounds simple and turns out not to be. Someone on a strict gluten-free diet who has an uneasy week wants to know one thing: did something get through? The usual answers are all indirect. You can consult how you feel. You can have blood drawn and look at antibodies. You can fill in an adherence questionnaire with a dietitian. All three are attempts to infer an exposure from its aftermath.

A more direct route exists. Gluten is not fully digested; particular fragments of it, called gluten immunogenic peptides or GIP, survive the gut and leave the body in stool and urine. Measure the fragments, and you are no longer inferring anything — you are detecting the gluten itself. The open question was how well that works at the small, accidental doses that matter in real life, and how it compares with the indirect methods people actually rely on.

The trial

A team led by Amy Russell and Jason Tye-Din ran a randomised, double-blind, placebo-controlled low-dose gluten challenge, published in Nutrients in 2024. Fifty-two people with coeliac disease, 36 of them women, median age 55, were split across five arms: placebo, 50 mg of gluten, 250 mg, 500 mg and 1000 mg.

The doses are the point. A slice of ordinary bread carries roughly 3 to 4 grams of gluten, so even the largest arm here is a fraction of a slice, and the smallest is closer to a crumb — the scale of a shared toaster or a sauce thickened with flour rather than a deliberate departure from the diet. Stool and urine were collected afterwards, alongside symptom scores, coeliac serology and a standard dietary adherence instrument.

What each method saw

Stool GIP found it. Every participant given 250 mg or more tested positive. At 50 mg, five of seven were detected. No one in the placebo arm tested positive, so the signal was not simply noise. Peptides peaked most often 12 to 36 hours after the dose, and stayed above the limit of quantification for a mean of about two days at 50 mg and about five days at 1000 mg.

Urine GIP did not. Across the intermittent low-dose exposures the trial was designed around, urine testing failed to identify a single one that stool testing caught. This is worth sitting with, because urine tests are the more convenient format and are marketed as such.

Blood tests did not. Transglutaminase-IgA and deamidated gluten peptide-IgG were negative in every participant at the end of the study — including those who had demonstrably eaten gluten. Serology is a good tool for its actual job. Catching a one-off low-dose exposure is not that job.

Symptoms did not. There was no correlation between the amount of gluten peptide recovered and the total symptom score on the coeliac patient-reported outcome instrument. The dietary adherence score fared no better: participants ranked by their pre-challenge stool GIP showed no significant difference in adherence score.

Why symptoms mislead in both directions

The absence of a correlation is easy to read too narrowly. It does not mean symptoms are imaginary or that nobody reacts to small amounts. It means the loudness of a reaction is a poor readout of the size of an exposure, and works badly in both directions. A quiet day is not proof of a clean day. A bad day is not proof that something got through, since the gut has many reasons to be unhappy.

That has a practical consequence for how people build habits. Anyone diagnosing their own diet from the feedback their body gives is running an experiment with an unreliable instrument, and will end up either over-trusting foods that happen not to produce a noticeable reaction or avoiding foods that coincided with a bad week. Neither error is visible from the inside.

What follows for the label in your hand

If the after-the-fact signals are weak, the before-the-fact ones carry more weight than people tend to give them. Checking what is in a product is not a substitute for a test; it is the step that reduces how often the question of an exposure arises at all. Reading an ingredient list closely, knowing which unfamiliar names indicate wheat, barley or rye, and noticing a "may contain" line are unglamorous habits, and this research is an argument that they matter more than the intuition people develop about their own tolerance.

It also reframes what a scanning app is for. It is a labelling tool, not a detector: it reads what a product declares and tells you what that declaration implies. It cannot see a shared fryer or an undeclared trace, and after this trial it is clear that neither can your own body with any reliability. The honest summary is that everyone involved — app, label, and person — is working with partial information, and that the partial information available before the meal is in better shape than the partial information available after it.

Limits worth naming

Fifty-two participants is a modest cohort, split five ways. The 50 mg finding rests on seven people. The challenge was a single controlled dose in a cookie, which is not the same as the drip of small exposures a real week delivers, and the study tested one commercial assay rather than the category as a whole. The core comparison — direct measurement outperforming inference — is nonetheless consistent with the wider literature on GIP monitoring, which has been building for several years.

Read the full paper: Russell AK, Lucas EC, Henneken LM, Pizzey CJ, Clarke D, Myleus A, Tye-Din JA. “Stool Gluten Peptide Detection Is Superior to Urinary Analysis, Coeliac Serology, Dietary Adherence Scores and Symptoms in the Detection of Intermittent Gluten Exposure in Coeliac Disease.” Nutrients 2024;16(2):279 — it is open access.

Common questions

Keep reading