Research Digest
The strain being rebuilt, and the trial that found nothing
Escherichia coli Nissle 1917 is the primary subject of seven papers here, and every one studies an engineered version rather than a probiotic. The best powered human trial reported no effect.
This batch brought in 303 papers across 37 strains, and those papers carry 188 strain links. Only 37 of the links are primary relevance, and that 37 is a floor rather than a count: another 77 were never scored at all, because their parent paper could not be summarized. Volume and evidence are different quantities, and they came apart clearly this time. One strain here is the primary subject of seven papers and every one of them studies an engineered version, while the best powered trial in the batch reported a null.
The trial randomised 428 infants in New Zealand to daily oral S. salivarius K12 or placebo from six months to twenty four months, and 368 of them contributed data on the primary outcome. Against placebo, it found no reduction in doctor recorded ear infections, and no difference in respiratory infections or antibiotic use either. The authors conclude that routine use for this purpose is not supported. This is the most informative trial in the batch precisely because it is the best powered of them and it answered in the negative. A null from a large trial narrows the field; ten small positive studies of varied design do not. It is also worth noting what the trial does not say. It tested one strain, one dose, one age range and one primary outcome, and a null there is not a statement about the organism in general.
The most engineered strain here is barely being studied as a probiotic
Escherichia coli Nissle 1917 is the primary subject of seven papers in this batch, and every one of them studies an engineered version. Three use it as a production host: one group blocked a competing tyrosine pathway to make it produce caffeic acid at 13.92 g/L in fed batch fermentation, another rebuilt it to synthesize HMB at 8.6 g/L, and a third turned it into a route to indican, an indigo precursor, at 1923 mg/L on crude glycerol. Three more give it a payload or a circuit, including outer membrane vesicles loaded with a drug and tested in colorectal tumour models in mice, a boron carrier for targeted radiation, and a calprotectin responsive circuit for mouse colitis. The seventh is a review of the same approach.
All of that is real work, and some of it is impressive. None of it is evidence that the organism does anything useful when swallowed as a supplement, which is the question this catalogue exists to track. Strain level paper counts cannot tell these apart. Nissle is third by link count in this batch, behind Akkermansia muciniphila and Faecalibacterium prausnitzii, so a reader ranking strains by activity would place it high and draw exactly the wrong conclusion about what that activity is. The gap between how much attention a strain receives and how much of it is aimed at its probiotic use is the thing a count cannot show.
Where a mechanism was actually pinned down
Most mechanism papers report that a strain changed something and that a pathway moved alongside it. One here went further. Working in zebrafish, a group showed that the Akkermansia muciniphila protein Amuc_1100 reduced diet induced liver fat and that it did so by binding an intestinal protein, 14-3-3b/a-A. The test that matters is the one they ran next: blocking that binding abolished the effect. That is a causal claim with a negative control attached, which is a different class of result from an observed association.
The same organism also drew the batch's one open disagreement. A perspective piece argues that reports of harm in injury models reflect mucosal damage and sampling artifacts rather than any toxicity of the organism, while an Aging Cell review covering the same literature lists strain variability and safety as the main limits on translating it into a therapy. Both are reading a real signal. Neither is in a position to settle it, and a catalogue entry that reported only one of them would be misleading.
Killed or alive, and the answer is not consistent
Three papers in this batch tested whether a heat treated preparation does the same work as a live one, and they split two to one. A heat inactivated Lactobacillus acidophilus LA-5 preparation matched the live strain in a mouse colitis model on colon length, splenomegaly and PANoptosis markers, and on ZBP1 and the IL-6 to IL-10 balance the authors report it doing better than the live probiotic, alongside raised butyrate. A heat inactivated Lacticaseibacillus paracasei 431 likewise matched its live form in a fermentation model, raising short chain fatty acids and shifting community composition. Against both, live Lactococcus lactis JCM 5805 extended healthspan in C. elegans through bacterial arginine metabolism, and the heat killed preparation did not work at all.
Read together they are not contradictory so much as a reminder that the question is badly posed. Whether killing the organism costs you the effect depends entirely on whether the effect runs through something the live cell has to keep doing. Where the active agent is a surface structure or an accumulated metabolite, inactivation may cost little. Where it depends on ongoing metabolism, it costs everything. Any general claim about paraprobiotics being equivalent is, on this evidence, premature.
What the human trials actually measured
Two trials of Lacticaseibacillus rhamnosus GG reported carefully bounded results. One found no independent probiotic effect on serum serotonin, with habitual short sleep also unlinked to higher levels. A dose comparison of encapsulated LGG reported gut symptoms and biomarkers stable at both doses, with a significant shift in gut microbiome composition at the higher dose and none at the lower one. That second one is a measurement of what the organism did, which is a different question from whether anyone felt better.
Three other trials did report outcomes a person would notice, and the honest complication is who deserves the credit. A randomized trial raised Faecalibacterium prausnitzii levels and eased flatulence symptoms, but the organism administered was Bifidobacterium longum CCFM1319; F. prausnitzii rose downstream of it. In 100 adolescent basketball players, a multi strain probiotic and prebiotic reduced immobilization related muscle loss and sped return to play while raising microbiome diversity. In 102 women, a twelve week multi strain preparation moved several hormones and also improved weight, waist circumference and anxiety and depression scores against placebo.
None of those three isolates a single catalogued strain. Two administered blends, and the third administered a different organism entirely and measured our strain as a mediator. That is not a reason to discount them, and it is the reason a per strain catalogue has to record them carefully: an entry that credited F. prausnitzii with easing symptoms would be reporting something the trial did not test. Where this batch is genuinely thin is not in outcomes but in attribution.
One paper here does report a contaminant interaction, and how it went unrecorded is the more useful finding. It describes an engineered strain reducing arsenite uptake in the gastrointestinal tract. It arrived without a usable abstract, failed summarization, and so carries no scored link, which means nothing in the catalogue surfaces it. The keyword scan that looks for contaminant mentions would not have caught it either: it matches arsenic, and this paper says arsenite. Across the scored links nothing else in the batch reported an interaction with microplastics, heavy metals or mycotoxins. That is an accurate statement about scored evidence and not about the literature, and the distinction is the point.