The bacteria in a person’s gut can change with dinner, medication, infection, age and the people they live with. Cognitive performance can change with sleep, education, health and the test being used. Put those two moving targets into the same observational study and an association may appear without revealing which direction the relationship runs, or whether a third factor produced both.
A study published in July 2026 tried to approach that problem through genetics. Its authors reported that inherited human variants used as proxies for the abundance of particular gut bacteria were also associated with cognitive performance. They then searched a large plasma-metabolite dataset for possible biochemical links.
The result is another clue in a growing human gut-brain literature. It is not evidence that a named bacterium makes people smarter, and it does not provide a diet or probiotic prescription. To understand the difference, it helps to look closely at what the researchers actually combined.
The study joined three datasets rather than testing one group
The open-access paper in Medicine used a method called two-sample Mendelian randomization. The microbiome side came from a genome-wide association study of 5,959 adults in Finland. Researchers in that source project had examined 7,967,866 human genetic variants in relation to 2,801 microbial taxa found through stool metagenomics.
The cognitive side came from a separate genome-wide meta-analysis of intelligence in 269,867 people. A third dataset covered 1,091 plasma metabolites and 309 metabolite ratios in 8,299 participants of European descent. The 2026 team worked with summary statistics from these resources. It did not recruit a fresh cohort, collect each person’s stool and blood, and administer a cognitive test.
The proposed chain looked like this: a human genetic variant is associated with the relative abundance of a microbial taxon; that microbial tendency is associated with a circulating metabolite; and the metabolite is associated with cognitive performance. Combining datasets gives the analysis considerable statistical reach, but every arrow is inferred across different groups of people.
Two highlighted bacterial signals sat among 24 suggestive results
For cognitive performance, the results section reported 11 taxa with positive and 13 with negative associations at an indicative threshold of P < .05. The authors highlighted Rhodanobacter and UBP9 on the positive side. A genetically predicted increase in Rhodanobacter was associated with an odds ratio of 1.113 for cognitive performance, with a 95 percent confidence interval from 1.001 to 1.237 and P = .047. UBP9 had an odds ratio of 1.080, a confidence interval from 1.012 to 1.152 and P = .019.
These numbers do not mean that stool samples from better-performing people contained 11.3 or 8 percent more bacteria. An odds ratio from the model describes the association between a standard-deviation change in a genetically predicted exposure and the outcome scale used in the source analysis. The unfamiliar taxonomic labels are also not products or probiotic strains that a consumer can meaningfully select.
There is a second reason for restraint. Testing thousands of taxa creates many opportunities for a nominal P value below .05 to appear by chance. The paper says false-discovery-rate correction was performed, yet its main results call the associations “indicative” and do not clearly state in the prose that the highlighted cognitive signals survived correction. Rhodanobacter’s confidence interval barely excludes the null value and its P value of .047 sits close to the conventional cutoff.
What Mendelian randomization can improve, and what it assumes
Ordinary observational evidence is vulnerable to reverse causation. Cognitive decline might alter diet, activity, medication use or independence, which could then alter the microbiome. Mendelian randomization uses inherited variants as instruments because genetic allocation occurs before the later-life outcome. Under the right conditions, this can reduce confounding and make the direction of an association more informative.
Those conditions are demanding. The variant must be reliably related to the microbial exposure. It must not be associated with factors that independently affect cognition. It must influence cognition through the proposed microbial route rather than through another biological pathway. That last failure is called horizontal pleiotropy, and it is especially difficult when human genes can influence immunity, metabolism, diet and the gut environment at once.
The authors used inverse-variance weighting as their primary model, checked several alternative estimators and screened for heterogeneity and pleiotropy. They retained instruments with F statistics above 10. These are sensible safeguards, but they cannot prove that every assumption holds. The STROBE-MR explanation and reporting guidance emphasizes instrument strength, overlapping biological pathways, population structure, multiple testing and the need for sensitivity analyses precisely because an MR estimate remains an inference, not random assignment of bacteria.
The metabolite result offers a possible bridge
The paper’s most specific proposed route involved 5α-pregnan-3β,20α-diol monosulfate, a steroid-related plasma metabolite. Higher genetically predicted levels of the metabolite were positively associated with cognitive performance, and the mediation model estimated that it accounted for 14.681 percent of the Rhodanobacter-cognition association.
That is useful as a hypothesis because metabolites are one plausible way for an intestinal ecosystem to communicate beyond the gut. Microbes can alter compounds that enter circulation, affect immune signalling, change gut barrier function or stimulate sensory pathways including the vagus nerve. This analysis directly modelled only the plasma-metabolite route. It did not observe the bacterium producing the steroid-related compound, track the compound into a brain or show neurons changing as a result.
The manuscript itself also switches once from “Rhodanobacter” to “Rhodobacter” in the mediation section, while the abstract and discussion use Rhodanobacter. That may be a typographical error, but the taxon-pathway label needs confirmation before anyone builds a biological story around it.
The finding joins several different kinds of gut-brain evidence
No single method can answer the entire question. Genetic studies are useful for direction and confounding. Longitudinal cohorts can show whether microbiome differences precede cognitive change. Randomized trials can test a defined intervention. Experiments in animals and cells can reveal mechanisms that would be difficult or unethical to isolate in people.
ScienceBlog recently covered mouse experiments connecting age-related gut changes, intestinal immune cells, vagal signalling and memory. Those experiments could perturb bacteria and pathways in ways this genetic analysis could not, but their relevance to humans remains to be established. An older ScienceBlog report described a proof-of-concept study in 36 women in which four weeks of a fermented milk product containing probiotics altered brain responses measured by functional MRI. That study was small and assessed a particular product, task and time window.
A 12-week randomized, double-blind trial in 63 healthy older adults later reported changes in cognitive and mood measures after supplementation with two specified Bifidobacterium strains. It provides experimental human evidence, but it cannot validate Rhodanobacter, UBP9 or the steroid-metabolite pathway. Microbiome effects are strain-specific, context-dependent and difficult to generalize from one age group or intervention to another.
Why this is a clue rather than a cognitive-health instruction
The microbiome genetic instruments were drawn from 5,959 adults and selected at a relaxed threshold of P < 1 × 10−5, used because few bacterial traits have variants meeting the usual genome-wide threshold. A separate 2026 HUNT study of host genetics and gut microbes shows why this remains difficult. Its discovery cohort included 12,652 people and replication involved Nordic cohorts of up to 21,976, yet many microbial genetic instruments were still weak or represented by a single signal. Across 546 species, mean SNP heritability was 6.8 percent.
Diet, drugs, household and geography may shape the microbiome more strongly than the inherited variants used as instruments. The present datasets were also predominantly European, limiting confidence that the same genetic proxies, microbes and effect estimates apply across ancestries and environments. All source data were from adults, so the study cannot establish how the relationship develops earlier in life.
Most importantly, the analysis did not test an intervention. There is no basis here for trying to increase either highlighted taxon, choosing a generic “brain probiotic” or assuming that changing the gut will improve a cognitive score. Future work needs to measure genetics, diet, medication, stool metagenomics, circulating metabolites and repeated cognition in the same people; replicate the taxon and metabolite signals in diverse cohorts; and then perturb a defined pathway safely.
ScienceBlog’s report on bacteria reaching mouse brains under specific experimental conditions illustrates the wider pattern. A route can be biologically real in an animal model without establishing its frequency, effect or clinical meaning in ordinary human life.
The 2026 genetic study narrows a vast field to testable candidates. It makes the gut-brain connection harder to dismiss as a metaphor, but it does not turn an association into a menu. For now, it gives researchers a shortlist, not readers a shopping list.