Your microbiome is an ecosystem, not a score

I spent the first half of my research life asking how bacteria in the human gut talk to the cells lining it. So I want to be careful here, because the thing I am about to say sounds like scepticism about my own field, and it is not. It is scepticism about a number.

The number in question

Send a stool sample to a company and you will get back a score. Diversity, usually, sometimes dressed up as a “gut health index”. It looks like a measurement. It has a decimal point. And there are four separate reasons not to trust it.

One. Diversity partly measures your bowel, not your health

Stool consistency is among the top-ranked covariates of gut community composition in population cohorts. A 2024 study in Nature Microbiology found that colonic transit time alone explains around six per cent of the variation in quantitative microbiome profiles and faecal pH around five per cent — more, in healthy adults, than diet does. Your diversity score is reading your plumbing as much as your ecology.

Two. One sample does not describe you

Twenty women, daily samples, six weeks. For seventy-eight per cent of bacterial genera, the day-to-day variation within a person exceeded the variation between people — in some cases a hundredfold. The authors put it plainly: single measurements do not estimate a person’s temporal average well, with a high risk of misclassification, because differences between patients and controls may fall inside normal temporal variation.

One sample, one number, one verdict. The verdict is noise with a decimal point.

Three. The companies do not agree with each other

In 2026 NIST did the experiment that had to be done: three identical aliquots of homogenised reference stool, sent to seven commercial testing companies. Biological variation eliminated by design. The result was major discrepancies both within and across providers, with variation between companies matching the biological variation between different people. Only three bacterial genera appeared consistently across all the analyses. One company returned contradictory verdicts on its own three identical samples: two “healthy”, one “unhealthy”.

Four. The favourable species largely do not have names yet

The largest diet–microbiome mapping to date, published in Nature in 2026 across more than thirty-four thousand people and validated in two intervention trials, found that most of the key health-associated species were previously uncharacterised. The unfavourable organisms tend to be the familiar, cultured, nameable ones. So a report that scores you on named species is scoring the minority of the signal that happens to have a name.

What is actually solid

Now the other side, because the mechanism is real and I worked on it.

• Butyrate is the primary energy substrate of colonocytes. Germ-free colonocytes are energy-starved and go into autophagy; butyrate rescues them.
• Butyrate oxidation consumes epithelial oxygen, which stabilises HIF-1α and drives the barrier gene programme. Remove the bacteria and you remove the oxygen sink.
• Short-chain fatty acids shape regulatory T-cell populations in the colon, largely through histone deacetylase inhibition.
• Commensal bacteria modulate host transcription factors directly — NF-κB, AP-1, PPARγ. That last one was my own work, and it is the point where the microbiome stops being a passenger and becomes a signal.

Two honest caveats on all of it. Most of this mechanism is mouse and cell-line work — germ-free animals, knockout lines, sodium butyrate applied to cultured cells. Extrapolating to a person eating lentils is an inference, not a measurement. And faecal short-chain fatty acid concentration is a poor biomarker of any of it: ninety to ninety-five per cent is absorbed in the colon, so what a test measures is the residue, not the production.

The example that makes the point

A 2025 trial with a hundred and forty-seven participants — four times the size of the Stanford trial it was replicating — compared a high-fibre diet, a high-fermented-food diet and a control. The fermented-food arm showed no diversity gain over control and no reduction in inflammation. The fibre arm was more interesting: diversity went down, while butyrogenic potential went up, transit time shortened and sleep quality improved.

Function improved. The score got worse. If you had been tracking the score, you would have stopped the thing that was working.

It is a preprint, and I will update this when the peer-reviewed version lands. But it is larger and better powered than the study it fails to replicate, and it changes what can honestly be said about fermented foods.

So what do you actually do

You stop optimising a number and start attending to a relationship: between what you eat, when you eat, how you move, what you are exposed to, what your gut is telling you, and the life you are actually living. Eat a wide variety of plants — the evidence supports variety; it does not support the number thirty, which came from a comparison bin in a self-selected citizen-science cohort and became a target through marketing rather than through science. Increase fibre slowly. Keep your rhythms. Take antibiotics when you need them, exactly as prescribed, and never skip a necessary treatment to protect your microbes.

And if something persists — blood, weight loss, fever, night-time symptoms — see a doctor. Not a testing company.

• Joos R, et al. — Examining the healthy human microbiome concept. Nature Reviews Microbiology 23:192–205, 2025.
• Procházková N, et al. — Gut physiology and environment explain variations in human gut microbiome composition and metabolism. Nature Microbiology 9:3210–3225, 2024.
• Vandeputte D, et al. — Temporal variability in quantitative human gut microbiome profiles. Nature Communications 12:6740, 2021.
• Servetas SL, et al. — Evaluating the analytical performance of direct-to-consumer gut microbiome testing services. Communications Biology 9:269, 2026.
• Asnicar F, et al. — Gut micro-organisms associated with health, nutrition and dietary interventions. Nature 650:450–458, 2026.
• Kelly CJ, et al. — Crosstalk between microbiota-derived short-chain fatty acids and intestinal epithelial HIF. Cell Host & Microbe 17:662–671, 2015.
• Nepelska M, et al. — Commensal gut bacteria modulate phosphorylation-dependent PPARγ transcriptional activity in human intestinal epithelial cells. Scientific Reports 7:43199, 2017.
• Sze MA, et al. — Fecal short-chain fatty acids are not predictive of colonic tumor status. mBio 10:e01454-19, 2019.
• Wastyk HC, et al. — Gut-microbiota-targeted diets modulate human immune status. Cell 184:4137–4153, 2021.
• van den Belt M, et al. — Distinct modulatory effects of high-fiber and fermented-food diets (GEEF trial). medRxiv preprint, 2025.