How Little Gluten Is Too Little? A 2026 Trial Puts a Number On It

Everyone managing celiac disease eventually asks the same question: how small does a gluten exposure have to be before it stops counting? An Australian trial published in Gastroenterology in 2026 answered it with actual doses — and the more useful finding may be the one about symptoms.

The international standard for calling a food "gluten-free" is 20 parts per million. That number is a regulatory compromise, and it has always rested on a thin evidence base: a handful of studies, mostly using gut biopsy as the endpoint, which is a blunt instrument for detecting a small response. A team based in Brisbane set out to redo the question with a far more sensitive readout, and to go much lower than previous challenge studies had gone.

What the trial did

This was a randomized, double-blind, placebo-controlled adaptive dose–response trial in 51 adults with biopsy-proven celiac disease who had been on a gluten-free diet for more than two years. Median age was 52 and 69% were female. Each participant underwent three oral challenges at four-week intervals — 153 challenges in total — receiving either placebo or a measured gluten dose somewhere between 1 mg and 1000 mg.

The readout was interleukin 2, a signalling protein that appears in the blood within hours of gluten ingestion in people with treated celiac disease. It reflects gluten-specific T cells waking up, and it is sensitive enough to register responses far too small to show up on a biopsy. The trial counted a participant as responding when their IL-2 at least doubled.

For scale: 1000 mg of gluten is roughly what a small bite of ordinary bread delivers. At the other end, 1 mg is a crumb.

What they found

The response was dose-dependent, and it did not vanish as neatly as the labelling standard might imply.

That makes 3 mg the lowest observed immune effect level and 2 mg the no observed immune effect level in this group. Note the untidiness in the middle of the table: 27% responded at 8 mg but only 17% at 13 mg, and nobody responded at 5 mg while some did at 3 mg. With a small number of people at each dose, that is what statistical noise looks like. The trend is real; the individual percentages should not be read too closely.

Fitting a curve to the whole dataset, the authors estimated the dose producing a response in half of patients at 111 mg (95% CI, 0–244 mg), in 10% of patients at 2.4 mg (95% CI, 0–5.3 mg), and in 5% at 0.8 mg (95% CI, 0–1.8 mg). The confidence intervals are wide, which the paper does not hide.

The finding about symptoms

Here is the result with the most practical reach. Symptom scores rose after challenges, but not beyond what placebo produced. People reported fatigue and mild digestive complaints after low doses — and reported them just as often after receiving nothing at all. The authors' conclusion is blunt: symptoms are unreliable at exposures below 1000 mg.

This cuts in both directions, and both are worth sitting with. A meal that produces no reaction is not thereby proven gluten-free. And a bad afternoon after eating out is not proof that the restaurant got it wrong. Self-monitoring by how you feel, which is what most people default to, is picking up something other than small gluten exposures.

What this does not say about gluten-free labels

The phrase "below current food-labeling thresholds" invites a misreading, so it is worth doing the arithmetic. Parts per million is a concentration, not an amount. A food sitting exactly at the 20 ppm limit contains about 0.56 mg of gluten per one-ounce serving. Reaching the 3 mg mark from that food alone would take roughly five to six ounces of it — and that assumes the product sits at the very top of the legal range. Independent testing programmes have found the large majority of gluten-free labelled products come in well under 5 ppm, not at 20.

So the trial does not show that compliant gluten-free food is dangerous. What it shows is narrower and still interesting: the dose at which the immune system starts to register something sits inside the same order of magnitude as the regulatory limit rather than comfortably above it, which is not what the standard's designers had strong evidence for either way.

Limitations

Fifty-one participants spread across nine dose levels plus placebo means few people per dose, which is why the confidence intervals are wide and the dose–response table has bumps in it. The cohort was adults on a long-established gluten-free diet at a single Australian centre; children and the newly diagnosed were not studied. Challenges were single doses, not the repeated daily trickle that real-world contamination looks like.

Most importantly, IL-2 is a biomarker. The researchers themselves note that the long-term clinical consequences of low-dose immune activation remain unknown. A measurable blip in a signalling protein is evidence that something happened; it is not evidence that damage accumulated.

The source

Daveson AJM, Craig E, Vitak A, Ware RS, Schafer J, Sehgal A, Bose U, Colgrave MJ, Hardy MY, Tye-Din JA, Anderson RP. A randomized double-blind, placebo-controlled dose–response study to assess the gluten threshold dose in celiac disease. Gastroenterology, 2026. The work was led from the Wesley Research Institute in Brisbane with collaborators at the University of Queensland, CSIRO Agriculture and Food, Edith Cowan University and the Walter and Eliza Hall Institute.

Read the paper (DOI: 10.1053/j.gastro.2026.03.011)  •  PubMed  •  Summary from the research institute

Common questions

Keep reading