Human evidence in gut–brain research

The microbiota–gut–brain axis is biologically plausible.

Experimental evidence has shown that microbial colonisation can influence stress physiology, immune development, microglial maturation, intestinal and blood–brain barrier biology, neurochemistry and behaviour. Microbial communities and metabolites can be manipulated in animals, communication pathways can be interrupted, and selected phenotypes can sometimes be transferred through the microbiota.^¹–⁴

The question in human research is different.

It is no longer simply whether communication between intestinal microbial activity and the nervous system is biologically possible.

It is whether variation in the human intestinal microbiota contributes meaningfully to differences in brain function, behaviour or mental health — and whether the available evidence can distinguish cause from consequence.

That is a much higher evidential threshold.

Human microbiome studies operate within a system that cannot be experimentally controlled in the way a laboratory model can. Diet, medication, age, sex, geography, socioeconomic conditions, sleep, stress, physical activity, gastrointestinal function and disease can all influence microbial ecology. Many of the same variables can also influence mental and neurological outcomes.^⁵–⁸

A difference in the microbiome between two groups is therefore a beginning.

It is not yet a mechanism.

Human association studies established that there was something to investigate

Much of the early human microbiota–gut–brain literature was necessarily observational.

Researchers compared faecal microbial communities between people with and without psychiatric, neurological or neurodevelopmental conditions and reported differences in microbial composition, diversity or particular taxa. Reviews and meta-analyses have since identified microbiome differences across conditions including depression, anxiety-related disorders, autism spectrum disorder, Parkinson’s disease and other neurological phenotypes.^⁹–¹³

These findings were scientifically important.

They demonstrated that clinical phenotypes could be associated with reproducible aspects of intestinal microbial ecology and provided hypotheses for mechanistic research.

But the results were also heterogeneous.

Different studies frequently identified different taxa. Some findings replicated across cohorts; others did not. Differences in sequencing methods, geography, diet, medication, diagnostic criteria, sample size and statistical analysis contributed to variation.^{⁵,⁶,¹⁰}–^¹³

This is not surprising.

The human microbiome itself is highly variable.

The problem begins when an association is translated into a statement that a particular microbiome causes the condition being studied.

Case–control design cannot establish direction

Consider a study comparing the faecal microbiota of people with major depression with that of healthy controls.

If microbial composition differs, several explanations remain possible.

The microbial difference may contribute to disease biology.

Depression may alter diet, appetite, sleep, physical activity or gastrointestinal function and thereby change the microbiota.

Antidepressant or other medication may influence microbial communities.

Systemic inflammation or metabolic differences may affect both the microbiota and mood.

The two groups may differ in lifestyle, age, body composition or other exposures.

Or several of these processes may operate simultaneously.^{⁵–⁸,¹⁴}

A cross-sectional study cannot determine which sequence occurred.

This is the central limitation of much human microbiome research: the microbiota is measured after the phenotype already exists.

The resulting association may be biologically real while its causal direction remains unresolved.

Medication is not a minor confounder

Medication deserves particular attention because it can be deeply entangled with clinical phenotype.

Large-scale studies have shown that numerous commonly used non-antibiotic drugs are associated with changes in gut-microbiome composition and function.^¹⁵,¹⁶ Psychiatric and neurological populations may use antidepressants, antipsychotics, anxiolytics, proton-pump inhibitors, laxatives, analgesics and metabolic medications, among others.

If medication use differs systematically between cases and controls, part of the observed microbial signature may reflect treatment rather than disease.

The problem is not solved simply by recording medication status.

Dose, duration, polypharmacy and interaction with diet or host physiology may all matter.

Medication can also lie within the causal pathway rather than functioning purely as a confounder. A treatment may improve symptoms while independently changing microbial ecology.

Human microbiome analysis therefore requires careful causal thinking before statistical adjustment begins.

Diet can produce the same interpretative problem

Diet strongly influences microbial composition and metabolism.^{¹⁷–¹⁹}

Mental-health conditions can also alter dietary behaviour.

Reduced appetite, emotional eating, food avoidance, socioeconomic constraints, fatigue and medication-associated appetite changes can all influence what reaches the intestinal microbial ecosystem.

A dietary pattern associated with depression may therefore contribute to a microbial signature that is subsequently described as a depression-associated microbiome.

That microbial signature may still matter biologically.

But the causal structure may be:

mental-health phenotype → dietary change → microbiome change

rather than:

microbiome change → mental-health phenotype.

Or the relationship may be reciprocal.

Human studies must be designed to distinguish these possibilities rather than choosing the most compelling narrative after the data are collected.

Taxonomic inconsistency does not necessarily mean functional inconsistency

One of the recurring challenges in human gut–brain research is limited agreement about which bacterial taxa characterise particular conditions.

This has sometimes been interpreted as evidence that the microbiome signal is too inconsistent to be meaningful.

That conclusion may also be too simple.

Different microbial communities can share functional capabilities. Taxonomic differences between populations may therefore coexist with similarities in microbial pathways or metabolites.^²⁰–²²

Functional analysis can sometimes reveal patterns that are less obvious at the taxonomic level.

This is one reason metagenomics and metabolomics have become increasingly important in human microbiota–gut–brain research.

But functional measurements do not eliminate the causal problem.

A metabolite associated with depression remains an association unless its origin, temporal relationship and biological action are demonstrated.

Moving from taxonomy to function improves biological resolution.

It does not automatically convert correlation into mechanism.

Large population studies change the scale of the evidence

Large population-based cohorts have strengthened human evidence by allowing microbiome associations to be examined in thousands rather than dozens of participants.

Valles-Colomer and colleagues analysed population-level microbiome data in relation to quality of life and depression and identified microbial features associated with mental-health-related phenotypes, including microbial metabolic potential relevant to neuroactive compounds.^²³

More recent large-scale work has continued to investigate microbial signatures associated with depressive symptoms and psychiatric phenotypes across independent cohorts.^²⁴

These studies offer several advantages.

Statistical power is greater.

Covariates can be examined more systematically.

Replication across populations becomes possible.

Rare microbial features can be studied more reliably.

But sample size does not change the logic of causal inference.

A very large cross-sectional association is still cross-sectional.

The estimate may be more precise.

Its direction is not necessarily more certain.

Longitudinal studies add time

Longitudinal design provides an important next step because microbial and host variables are measured repeatedly.

If microbial change consistently precedes a clinical change, some causal explanations become more plausible and others less so.

Repeated sampling can also distinguish stable individual characteristics from transient fluctuations.^²⁵,²⁶

This principle was central to the design of our Meals, Microbiota and Mental Health in Children and Adolescents study.

The MMM-Study was designed as an observational longitudinal case–control study integrating dietary intake, intestinal microbiota, intestinal-permeability-related measures and mental-health phenotypes.^²⁷ Rather than treating the microbiota as an isolated exposure, the design recognises that food, microbial ecology, barrier biology and mental health can change together over time.

Longitudinal observation still does not establish causality.

But it introduces something cross-sectional studies cannot provide:

temporal structure.

That is an essential component of causal inference.

Human barrier evidence requires particular caution

Intestinal permeability is frequently proposed as an intermediate mechanism connecting microbial alterations with systemic or neural outcomes.

The hypothesis is plausible.

Changes in barrier regulation could alter host exposure to luminal antigens, microbial products or inflammatory signals, potentially influencing systemic physiology.^²⁸,²⁹

But human measurement is difficult.

Functional permeability tests, epithelial-injury markers, microbial-translocation markers and zonulin-related measurements do not measure the same biological property.^³⁰,³¹

This problem was evident in our systematic review and meta-analysis of zonulin-dependent intestinal permeability in children diagnosed with mental disorders.^³²

The available studies suggested differences in zonulin-related measurements in some populations, but methodological heterogeneity and the nature of the measurements limited interpretation.

A higher reported zonulin concentration cannot, by itself, demonstrate the complete sequence:

microbiota alteration → zonulin-pathway activation → increased intestinal permeability → systemic inflammation → altered brain function → mental-health phenotype.

Every arrow is a separate claim.

Every claim requires evidence.

Human metabolomics can move closer to mechanism

Metabolomics provides another route towards functional human evidence.

Rather than asking only which microorganisms differ, investigators can examine microbial-associated metabolites in stool, plasma or other biological samples.^³³–³⁵

This can help identify candidate pathways connecting microbial activity with host physiology.

Tryptophan metabolism is one important example. Alterations in microbial indole pathways, host kynurenine metabolism and serotonin-related biology have all been investigated in relation to brain function and psychiatric phenotypes.^³⁶,³⁷

Short-chain fatty acids and bile-acid metabolites provide other potential routes.^³⁴,³⁸

But metabolomics introduces its own interpretative boundaries.

A circulating metabolite may be shaped by diet, microbial metabolism, intestinal absorption, hepatic transformation, renal clearance and host disease.

Its detection does not prove microbial origin.

Its association with a phenotype does not prove biological mediation.

Functional evidence is closer to mechanism than taxonomy alone.

It is still not the mechanism itself.

Intervention studies test whether the system can be changed

Intervention trials provide stronger evidence than observational association when they deliberately alter one component of the microbiota–gut–brain system.

Probiotics, prebiotics, synbiotics and dietary interventions have all been studied in relation to mood, stress, cognition and other psychological outcomes.^{³⁹–⁴²}

Some trials report beneficial effects.

Others find little or no effect.

Meta-analyses suggest that selected interventions may influence depressive or anxiety symptoms in some populations, but heterogeneity remains substantial. Differences in strains, dose, duration, baseline microbiota, clinical population and outcome measures complicate interpretation.^{³⁹–⁴²}

The term psychobiotic emerged from this literature to describe microorganisms or microbiota-directed interventions with potential mental-health effects.^⁴³

The concept is useful when it generates testable hypotheses.

It becomes less useful when it implies that all probiotics have predictable psychological effects.

Microbial interventions are biologically specific.

A result obtained with one strain cannot automatically be generalised to another.

An intervention effect still does not identify the pathway

Even a successful randomised controlled trial may leave mechanism unresolved.

Suppose a probiotic intervention improves a measure of perceived stress.

Several explanations remain possible.

The microorganism may alter microbial-community function.

It may produce a metabolite.

It may influence intestinal immune activity.

It may alter gastrointestinal symptoms that themselves affect wellbeing.

It may act through a vagal or endocrine route.

Or the observed effect may involve several pathways.

A clinical effect demonstrates that the intervention influenced the measured outcome under the conditions of the trial.

It does not automatically establish the microbiota–gut–brain mechanism responsible.

Mechanistic substudies therefore matter.

Faecal microbiota transfer in humans is a different evidential category

Faecal microbiota transplantation provides strong therapeutic evidence in recurrent Clostridioides difficile infection.

Its role in neurological or psychiatric conditions is far less established.^{⁴⁴,⁴⁵}

Small studies and early trials have explored FMT or microbiota-transfer approaches in conditions involving neurodevelopmental, neurological or psychiatric symptoms, but sample sizes, study designs and endpoints remain insufficient for broad clinical conclusions.^{⁴⁴–⁴⁶}

Human FMT also differs conceptually from human-to-mouse transfer experiments.

In an animal model, donor microbiota can be transferred into highly controlled recipients to test transmissibility of a phenotype.

In a human intervention, the recipient already has a complex microbiome, medical history, diet, immune system and environmental context.

Engraftment and response vary.

The two forms of evidence should not be conflated.

Human genetics can help test causal hypotheses

Observational epidemiology is not the only route towards causal inference.

Host genetics and Mendelian-randomisation approaches have increasingly been used to test whether genetically predicted microbial features are associated with disease outcomes.^{⁴⁷,⁴⁸}

These approaches can reduce some forms of confounding and reverse causation.

But microbiome Mendelian randomisation is technically challenging.

Microbial traits often have relatively weak genetic instruments.

Taxonomic definitions differ between datasets.

Population structure and pleiotropy can influence results.

And a genetically associated microbial taxon remains an imperfect proxy for microbial function.^{⁴⁷,⁴⁸}

Genetic approaches therefore add another layer of evidence.

They do not provide an automatic shortcut to causality.

Multi-omics can connect the levels

One of the most promising directions in human microbiota–gut–brain research is integration across biological levels.

Microbiome composition can be combined with metagenomics, metabolomics, immune markers, barrier measurements, neuroimaging and detailed clinical phenotyping.^{²⁴,²⁵,³³}

This can reveal relationships that would remain invisible within a single dataset.

But multi-omics also creates a risk of narrative overfitting.

When thousands of microbial, metabolic and host variables are measured simultaneously, many correlations can be found.

A coherent pathway reconstructed after the analysis may be biologically plausible without having been prospectively tested.

Replication, validation cohorts and mechanistic follow-up remain essential.

More layers of association do not automatically equal mechanism.

Human neuroimaging offers another bridge

Neuroimaging has been used to examine relationships between gut microbial features and brain structure, functional connectivity or responses to emotional stimuli.^{⁴⁹–⁵¹}

This is an intriguing development because it moves the neural outcome beyond self-reported symptoms.

But imaging findings are still associations unless study design supports stronger inference.

A microbiome feature associated with functional connectivity does not demonstrate that the microorganism changed the brain.

Brain state, diet, stress, medication or host physiology may influence both variables.

Neuroimaging increases phenotypic resolution.

It does not eliminate causal ambiguity.

Replication matters more than novelty

Microbiome research generates high-dimensional datasets in which exploratory findings are relatively easy to produce.

The more difficult task is replication.

Human gut–brain studies are particularly vulnerable to cohort-specific effects because both microbiome and psychological phenotypes are influenced by population characteristics.^⁵–⁸

A microbial signature discovered in one cohort should therefore be tested in another population using compatible methods.

Replication across geography, sequencing platforms and analytical pipelines provides stronger evidence than increasingly complex analysis of a single dataset.

This is especially important before microbial biomarkers are proposed for clinical use.

The clinically relevant question is not whether the axis exists

At this stage, asking whether a microbiota–gut–brain axis exists is no longer especially useful.

There is extensive experimental evidence for communication between microbial, intestinal and neural systems.^¹–⁴

The clinically relevant questions are more specific.

Which microbial functions influence which human outcomes?

Through which pathways?

At what magnitude?

During which developmental or disease context?

Can the signal be reproduced?

Can it be modified safely?

Does modifying it improve an outcome that matters to patients?

And can the effect be separated from diet, medication, host physiology and other environmental influences?

These questions are harder than demonstrating association.

They are also where translation begins.

Human evidence should become more specific as the field matures

Early microbiota–gut–brain research needed broad hypotheses.

The field first had to establish that microbial ecology could plausibly influence neural biology.

That foundation now exists.

As the science matures, claims should become narrower rather than broader.

Not:

the microbiome influences mental health.

But:

a defined microbial function alters a measurable host pathway, under specified conditions, in a population in which the effect has been replicated.

Not:

increased intestinal permeability links the gut to the brain.

But:

a validated permeability measure changes before a defined systemic signal, and that signal predicts or mediates a subsequent neural outcome.

Specificity is not a retreat from the microbiota–gut–brain concept.

It is evidence that the field is becoming more rigorous.

From association to human mechanism

Human research sits at the most difficult point of microbiota–gut–brain translation.

The system cannot be fully controlled.

The exposures are complex.

The outcomes are heterogeneous.

And many variables influence both microbiota and brain simultaneously.

Yet human research is also where the biological hypothesis ultimately has to earn its clinical meaning.

Cross-sectional studies identify associations.

Large cohorts test their robustness.

Longitudinal studies establish temporal structure.

Interventions test modifiability.

Metabolomics and multi-omics identify candidate pathways.

Neuroimaging refines neural phenotypes.

Genetic approaches probe causal direction.

Mechanistic experiments test the links that human studies cannot isolate.

No single design completes the evidence chain.

The strength lies in convergence across them.

The central question is therefore not whether a human microbiome finding is interesting. It is how much of the path from microbial observation to human brain outcome has actually been demonstrated.