Research Digest
Three randomised results arrived, and none of them has been through peer review
Two are preprints and one is a doctoral thesis. Two of the three report no difference between the probiotic and the placebo. A strain page counts papers, so this is the kind of thing a count cannot show you.
This sweep queried all 508 active strains over a window reaching back to June and brought in 195 papers, every one published in 2026. It created 904 link rows; off-context pruning removed 774 of them and three alias-pollution rows accounted for the rest, leaving 127 links across 31 strains, of which 72 are scored primary or secondary. Those 72 links sit on 67 distinct papers: 22 in vitro, 20 animal, 15 reviews, 7 observational, and 3 carrying randomised evidence. Not one of those three has been through peer review.
A randomised, double-blind, placebo-controlled trial in 34 patients with symptomatic uncomplicated diverticular disease tested a synbiotic containing Lacticaseibacillus paracasei DG plus inulin against placebo, both on a high-fibre diet. Gastrointestinal symptoms improved in both arms over twelve weeks, with no difference between them, and quality of life did not change significantly in either. That is a null result from the strongest design here, and it is sitting on Research Square rather than in a journal. Both halves of that sentence matter. A null trial is a real contribution, and a reader deciding what to make of a strain deserves to know that the trial has not yet been refereed.
The other two, and what a paper count hides
The second randomised paper is a medRxiv preprint that gave Lacticaseibacillus rhamnosus HN001 to adults aged 60 to 80, 55 women and 28 men, for six weeks, in two formulations, with brainstem imaging before and after. Connectivity measures moved over the six weeks, and the paper reports no differences between the groups. The third is a doctoral thesis describing an open-label randomised trial linked to Lacticaseibacillus rhamnosus GG, with 30 patients enrolled and 25 completing. We are describing its design and not its outcome, which is a deliberate omission rather than an oversight: restating a human efficacy result for a probiotic, on the property of a company that sells probiotics, is a product claim by proxy whatever the source. The design is the part a reader can use anyway: open-label, 30 enrolled and 25 completing, and not yet refereed.
None of this is a complaint about the authors. Preprints and theses are ordinary scientific outputs, and a field where two of three trials report no difference is a field being honest with itself. The point is narrower and it is about this index. A strain page shows a number of papers. It does not show that the one randomised paper behind that number has not been refereed, and of the 67 papers scored in this batch, 10 are preprints, 2 are theses, 2 are datasets and 1 is a conference abstract. Counting links is a poor proxy for counting evidence, which is why these pages report the design mix.
Nissle keeps being used as a chassis, for the third batch running
Escherichia coli Nissle 1917 took 10 scored links, 9 of them primary, and eight of those nine study the organism modified or decorated rather than as anything a person can buy. It was given urate oxidase modules from its own plasmid in one study, engineered to home to infected bone and secrete an enzyme and a peptide in another, loaded with chitosan nanoparticles carrying a neurotransmitter, and made to express an immune signalling protein alongside an Akkermansia protein. The ninth primary link, a review of the unmodified organism's own biology, is also the one that raises the strain's pks island as an open safety question. The 2026-10-12 batch split 11 engineered against 6 unmodified on this strain and the 2026-10-25 batch split 6 of 7 the same way, so a tall paper count on this strain page is largely counting a different activity from the one a reader is shopping for.
Two of the eight are the only contaminant-tagged links in the entire batch, and they are the ones closest to this catalogue's own subject. A zinc-loaded Nissle under a hyaluronic acid shell was given to zebrafish exposed to polystyrene nanoplastics, and restored behavioural, immune, redox and neurochemical readouts further than its single-component controls did. A Nissle displaying Tff1 on its surface raised faecal copper excretion and lowered copper in liver and urine in a mouse model of copper poisoning. Both are scored primary with a host-context setting, and neither moved a mark on the strain page: Nissle already carried host-context evidence in all three contaminant categories, and no confidence level moved anywhere among 5,984 rollup rows this run.
What neither paper is, is evidence about a product. One organism was wrapped in a synthetic polymer and filled with a mineral; the other was given a gene it does not have. Both ran in animals. Read them as demonstrations that the chemistry can be made to work inside something living, which is genuinely interesting, and not as information about anything on a shelf.
One organism, three papers, two directions
Faecalibacterium prausnitzii took 9 scored links, and three of them study the organism itself rather than measuring its abundance alongside something else. Two point one way. An article in mSystems found it depleted in the faeces of patients with sepsis, and reports that giving it orally improved survival and reduced organ injury in septic mice, with one of its metabolites reproducing the protection. A doctoral thesis reports that butyrate, which this organism produces from dietary fibre, suppressed the inflamed state of immune cells by inhibiting an enzyme called HDAC3, across human biopsies, organoids and immune cell cultures.
The third points the other way. A conference abstract in the European Respiratory Journal reports that in lung adenocarcinoma samples the organism was enriched relative to normal lung tissue, and that in cell culture and a xenograft model it enhanced malignant phenotypes and tumour growth. A conference abstract is the thinnest kind of record in this batch and it has not been refereed either, so the right weight to give it is small. But the right weight is not zero, and a strain index that files the first two papers and quietly drops the third is not a research index. An organism that is protective in one tissue and one disease model is not thereby protective everywhere, and the same batch that supplies the encouraging reading supplies the caution.
The batch's largest share, and its own reviews arguing for restraint
Akkermansia muciniphila took 16 of the 72 scored links, more than any other strain, and the mix is mostly animal studies and reviews. The most useful two are reviews that argue against their own subject's momentum. One survey of the organism across many conditions notes that in some host contexts, or alongside certain pathogens, it may worsen rather than ease disease. A critical review of 50 publications on its extracellular vesicles reaches the conclusion that the evidence does not establish either efficacy or safety of isolated vesicles in patients, and that their value depends on what they add beyond whole bacteria or purified components. That is the review literature of a popular organism telling its readers to slow down, and a catalogue organised by strain should be repeating it rather than counting past it.
A third paper shows where the field is actually going. In a mouse study, Amuc_1100, a single protein from this organism's outer membrane, reproduced several of the protective effects of the live and the pasteurized bacterium. The interesting unit of analysis is drifting from the organism to its parts, which is the same drift the Nissle section describes from the other end. One review in this batch explains why that drift has commercial force: most of these next-generation strains are strict anaerobes that readily lose viability on contact with oxygen, processing stress or digestive fluids. A protein does not have that problem. Neither does a gene in a different organism. Whether either still does what the bacterium did is the question the next few batches will be answering.