Resources
The evidence, and where it stops.
Three domains I have actually worked in, summarised honestly — including the parts that are contested, the parts that are mouse work, and the popular claims that turned out to be wrong.
8peer-reviewed journal articles
3European Commission reports
2AOP papers with the OECD network
1patent on a bacterial transporter
Domain one
The microbiome
My first research life. How commensal bacteria and their metabolites speak to the cells lining your gut — and why almost everything sold as a microbiome “score” is measuring the wrong thing.
What is solidly established
- Butyrate is the primary fuel of colonocytes. Germ-free colonocytes are energy-starved; butyrate rescues them.
- Butyrate oxidation consumes epithelial oxygen, stabilises HIF-1α and drives the barrier gene programme.
- Short-chain fatty acids shape regulatory T-cell populations in the colon, largely through HDAC inhibition.
- Commensal bacteria modulate host transcription factors directly — NF-κB, AP-1, PPARγ. This is the work I published.
Evidence: strong mechanistically — but mostly in cells and mice
What is weaker than you have been told
- “Diversity equals health” is a confounded population-level statistic. Colonic transit time and stool consistency explain more variance than diet does.
- One stool sample does not estimate your own average. For 78% of genera, your day-to-day variation exceeds the variation between people.
- Faecal short-chain fatty acids measure what was not absorbed. Any test reporting your “butyrate level” is reporting a residue.
- “Thirty plants a week” comes from a comparison bin in a self-selected citizen-science cohort, not from a trial. Eat variety; the number 30 is marketing.
Evidence: weak, contested, or over-extended in popular use
Photo to addDrop in a figure from one of your papers — the epithelial signalling schematic works well here.
My own papers in this domain
- Liuu S, Nepelska M, et al. — Identification of a muropeptide precursor transporter from gut microbiota and its role in preventing intestinal inflammation. PNAS 120, 2023.
- Nepelska M, et al. — Commensal gut bacteria modulate phosphorylation-dependent PPARγ transcriptional activity in human intestinal epithelial cells. Scientific Reports 7:43199, 2017.
- de Wouters T, Ledue F, Nepelska M, et al. — A robust and adaptable high throughput screening method to study host–microbiota interactions. PLoS ONE 9:e105598, 2014.
- Nepelska M, et al. — Butyrate produced by commensal bacteria potentiates phorbol esters induced AP-1 response in human intestinal epithelial cells. PLoS ONE 7:e52869, 2012.
- Lakhdari O, et al. (incl. Nepelska M) — Identification of NF-κB modulation capabilities within human intestinal commensal bacteria. J Biomed Biotechnol 2011:282356.
Domain two
Endocrine disruptors
Six years inside the European system that decides what counts as one. Here is the actual definition, the actual mechanisms, and the actual disagreement.
The definition matters
The WHO/IPCS definition, made legally operative in the EU: an exogenous substance that alters endocrine function and consequently causes adverse health effects in an intact organism, its progeny or sub-populations.
Three elements must all hold: an adverse effect, an endocrine mode of action, and a biologically plausible link between them. “Endocrine active” is not “endocrine disruptor”. Many foods are endocrine active.
How they act
The regulatory battery is built around four modalities — estrogenic, androgenic, thyroid, steroidogenic. Beyond receptor agonism and antagonism: altered hormone synthesis, transport, metabolism, receptor expression, and epigenetic effects on hormone-responsive genes.
The honest gap: retinoid signalling, PPAR and metabolic pathways, the hypothalamic–pituitary axes and glucocorticoid signalling are largely outside the validated test set. Absence of evidence there reflects absence of tests, not absence of effects.
Where it is genuinely contested
Whether dose–response can be non-monotonic, so that a high-dose no-effect finding does not license extrapolation downward. After twenty-five years there is still no blinded multi-laboratory ring trial settling it.
In 2023 EFSA cut the tolerable daily intake for bisphenol A by a factor of 20,000. The German BfR and the EMA formally dissented. That disagreement is between regulators, not between science and industry.
Where the EU stands, 2026
Sectoral criteria in force since 2018 for biocides and plant protection products. Endocrine disruption became a horizontal CLP hazard class in 2023, applying to new substances from May 2025 and to mixtures from May 2026.
In its November 2025 review the Commission judged the criteria robust, while naming its own constraints: too few qualified toxicologists in Member States, too few certified testing facilities, and reviews running past three years.
Evidence: the framework is settled — its application is not
Photo to addA figure from your endocrine disruptor work, or the EATS modality diagram.
Domain three
Regulatory toxicology and Adverse Outcome Pathways
The tool that made everything else possible. A causal map from a molecular trigger to an observable harm, where every arrow has to be defended on its own.
Position paper
AOP development: strategies and principles
Co-authored with the OECD-facing network. Sets out how a pathway is assembled: the molecular initiating event, the key events between, and the relationships that link them — each weighted for biological plausibility, empirical support and essentiality.
The core argument: if the causal chain is solidly established, a risk can be assessed by measuring an early event, without reproducing the whole experiment on an animal.
Position paper
AOP development: best practices
The companion paper. What separates a usable pathway from a plausible story: explicit confidence per link, declared uncertainty, and the discipline of removing an arrow you cannot defend even when it made the narrative cleaner.
Position paper
PPARα activation and reproductive toxicity
First author. A pathway from PPARα activation in utero to impaired fertility, developed for the assessment of endocrine disruptors and reproductive toxicants — and the paper where the two halves of my record meet.
The honest limitation
Most AOPs remain qualitative. Quantitative pathways — the kind that would let you predict how big an outcome will be from how big a key event is, and which are exactly what you would need to settle the low-dose question — are still the exception. AOPs structure the argument. They do not yet settle it.
Evidence: a mature framework, an immature quantification
Papers and reports
- Nepelska M, Odum J, Munn S — Adverse Outcome Pathway: PPARα activation and reproductive toxicity. Applied In Vitro Toxicology 3:234–249, 2017.
- Villeneuve DL, et al. (incl. Nepelska M) — AOP development I: strategies and principles. Toxicological Sciences, 2014.
- Villeneuve DL, et al. (incl. Nepelska M) — AOP development II: best practices. Toxicological Sciences, 2014.
- Worth A, et al. (incl. Nepelska M) — Alternative methods for regulatory toxicology: a state-of-the-art review. JRC Science and Policy Report, 2014.
What comes next
From adverse outcomes to empowered pathways.
Thirty years of collective work produced an infrastructure for mapping how a living system degrades. There is no equivalent for mapping how it recovers. That asymmetry is the research programme I am building, and it joins the two halves of my record: the microbiome as a modifier of the pathway, and the pathway as a discipline for talking about recovery without hand-waving.
Same causal logic. Same requirement to declare confidence link by link. Applied to the climb back rather than the fall.
This is stated as a direction of work, not as a validated method. When it becomes one, the evidence will be here.
In video
Short formats
The same ideas, explained in a few minutes. Nutrition, the microbiome, and how the two interact.
YouTubePaste the embed code here — nutrition and the microbiome.
YouTubePaste the embed code here — how food, rhythm and gut bacteria interact.
YouTubePaste the embed code here — what a microbiome test can and cannot tell you.
YouTubePaste the embed code here — free slot.
Next step
Three pilot teams this year.
Twenty-five minutes to check whether this format answers a real problem at your organisation. If it doesn’t, I’ll say so.