Research Digest
The papers only one source had
Ninety one of this batch's 266 papers came from PubMed alone. Until this week none of them could enter the index at all, and the strain that lost the most to that gap is the one this catalogue tracks most closely.
This batch brought in 266 papers carrying 195 strain links across 42 strains. Of those links 114 are scored primary or secondary, spread over 30 strains. What makes the batch unusual is not its size but where it came from: 91 of the papers were indexed by PubMed and not by OpenAlex, and until this week not one of them could be stored. A defect in how PubMed records were parsed made every PubMed-only paper fail silently on write from May onward, so the index has never held a single one. That is a finding about our own method rather than about the literature, and it changes how the rest of this batch should be read.
Fifty three adults with a history of repeated sore throat were randomised to Streptococcus salivarius K12 lozenges either daily or once every seven days, over a fourteen day course, with colonisation measured by PCR on self-taken swabs at six timepoints. Daily dosing colonised more people and held longer: on day 7, 85.7% of the daily group carried it against 13.6% of the once-every-seven-days group. By day 21 one participant in the entire trial still did. The authors' conclusion is the part worth keeping: colonisation was not maintained once dosing stopped. This was a feasibility trial, sized to test whether the regimen works as a procedure rather than whether it helps anyone, and it reports no clinical outcome at all. What it does report is a direct measurement of something usually assumed rather than checked, which is how long a swallowed organism persists. On this evidence, for this strain, it persists while you keep taking it.
What the gap was hiding
Forty four of this batch's 114 scored links sit on papers PubMed alone indexed. Akkermansia muciniphila is the clearest case: it holds 34 scored links, the most of any strain here, and 17 of them come from papers OpenAlex did not have. Exactly half of what this batch tells us about the strain was invisible to the method that built the rest of the index.
It would be wrong to read that as the corpus missing half its literature. OpenAlex alone still supplied 138 papers here against PubMed's 91, and 37 came from both. Across the whole index, 48.7% of papers carry both identifiers, and that figure is itself flattered by the defect, since until this week every paper PubMed alone knew about was refused. The gap was specific rather than proportional. It fell on recently indexed records, which is precisely where a sweep windowed on publication date is looking, so the loss concentrated in new work rather than spreading evenly across the archive. Nothing in the pipeline reported it, because a paper that fails to save leaves no row to count and the failure was never logged. Counts cannot show you what was never written.
The strain with the most evidence also drew the most complications
Akkermansia leads this batch on volume, and three of the papers it gained argue against reading that volume as a simple endorsement. In gnotobiotic mice, colonising with either A. muciniphila or Bacteroides thetaiotaomicron alone produced no constipation, while colonising with both together did, through a sulfatase-dependent route for degrading mucin. A commentary asks whether antibiotic-induced mutants of the organism survive better while interacting less usefully with the host, and proposes that as a link worth testing against rising metabolic disease rates. A third reports that across independent vaginal metagenomic datasets, samples positive for A. muciniphila carried higher group B Streptococcus abundance, and that the organism promoted that species' attachment to vaginal cells in a laboratory model.
None of those says the organism is harmful, and the third reports that continued treatment reduced the other species' persistence in mice, which cuts the other way. What they say jointly is that its effects depend on what else is present and on where it is. That is the one thing a per-strain catalogue is least equipped to show, because it files evidence under a strain and the finding lives in a combination.
One trial named its outcome in advance, and it came back null
A randomised placebo-controlled pilot gave heat-killed Bifidobacterium breve B-3 to patients with sarcopenia and found no improvement in knee extension strength, its primary outcome. The paper notes that exploratory, unadjusted analyses suggested something for muscle mass and endocrine markers, and the adjectives matter: exploratory and unadjusted is a reason to run another trial, not a result. This is the most informative human result in the batch for the same reason the lead study is. It set out to measure one named thing, measured it, and reported that it did not move.
The batch's other human null is weaker evidence, and it is worth being precise about why. A randomised trial of orange juice fortified with probiotics and vitamin D3 reported no significant improvement in glucose metabolism against unfortified juice, but that glucose finding is itself a secondary-outcomes analysis rather than the question the trial was designed and powered to answer. The same standard that discounts the B-3 trial's favourable exploratory numbers has to discount this trial's unfavourable secondary ones. A secondary analysis narrows nothing in either direction, and treating a null as more trustworthy than a positive simply because it is a null would be the same error wearing different clothes.
Where attribution got harder
A trial randomised 124 children to either Limosilactobacillus reuteri DSM 17938 or Lacticaseibacillus rhamnosus GG and compared them directly. Pain frequency fell in both groups, and the difference between the strains was not significant. That is worth recording precisely because this catalogue is built on strain-level distinctions: here two of its better documented strains were put head to head and separated by nothing measurable.
The contaminant links, and what each one is not
Four papers in this batch carry a contaminant tag, between them producing five strain links. They are worth listing individually, because the category name is the same in each case and the evidence behind it is not.
The one closest to this catalogue's subject is a narrative review of dietary interventions against micro- and nanoplastic toxicity, which named two strains this index already tracks, Lacticaseibacillus paracasei DT66 and Lactiplantibacillus plantarum DT88. Each page now lists three papers instead of two, and the microplastic mark on each is unchanged. The page moved and the mark did not, and that is the point. Both strains already carried that mark, earned in May from an animal study in which each was the paper's central subject. This review names them in passing, so both new links are scored secondary, and the contaminant categories count primary evidence only. A narrative review that surveys a field does not add evidence to it, and the mark correctly does not move.
The review's own verdict is the part to carry forward: the work it surveys is preclinical, and it reports no supporting human data. That is worth holding onto next to the mark those two strains do carry, which rests on one animal study.
The second is chemical rather than particulate. Bisphenol F, a bisphenol A substitute the authors detected in 90.5% of 285 human urine samples, was converted by gut bacteria into N-acetylputrescine, which damaged the intestinal barrier and produced cardiac hypertrophy in mice; the effects depended on gut microbiota being present, and supplementing A. muciniphila or its metabolite tryptophol reduced them. It is filed under chemical interaction rather than microplastic interaction because leached compounds are not particles, a distinction the category names are built to keep.
The third is a heavy metal result worth naming for what it is not: an engineered E. coli Nissle 1917 displaying lead-binding curli nanofibers bound up to 70 mg of lead per gram of biomass and lowered blood lead by roughly 30% in mice. That is a synthetic biology result about a modified organism, not evidence about any probiotic taken as a supplement. The fourth is thinner still: an in vitro test finding that methyl methacrylate at residual denture-monomer concentrations did not significantly inhibit growth of Streptococcus salivarius M18, which is a bench observation about a dental material and carries an unclear setting.
One of the four did move a mark: the denture-monomer result took Streptococcus salivarius M18 from no mycotoxin and chemical evidence to environmental, which caps at the weakest grade and lights nothing on the card, because a bench test is not a finding about a body. Two of the four are engineered or in vitro, one is a review reporting no human data, and one is an animal study of a chemical. That is the honest shape of the contaminant evidence this batch added, and it is the reason the index records study design and setting beside the count rather than letting a tag stand on its own.