The Ultimate Guide to Optimising Brain Function Through Microbiome Modulation
Twenty years ago, telling a neurologist that the state of someone's colon might shape the state of their memory would have earned you a polite silence. Today, the microbiota–gut–brain axis has its own 137-page review in Physiological Reviews, and the question has shifted from whether the gut talks to the brain to which signals it uses, and which of those we can actually move (Cryan 2019).
That shift changes what we measure. For years the field chased bacterial names — more of this species, less of that one — and the results refused to replicate. Compare the microbiome studies in depression, Parkinson's disease, autism, ME/CFS and Alzheimer's disease side by side and the taxa lists disagree, sometimes flatly. Look instead at what those bacteria do, and a single pattern comes into focus across all of them: the microbes that make butyrate and indole-3-propionic acid are depleted, the microbes that make inflammatory endotoxin and barrier-degrading metabolites are enriched, and the two gatekeepers standing between the gut and the brain — the intestinal barrier and the blood–brain barrier — are under strain.
One signature, not nine fingerprints. That is the through-line here, and it is why the clinical work is more tractable than it first appears: you are not trying to fix nine different microbiomes, you are trying to shift one balance. What follows is in three parts — why the gut has any say in brain function at all, what the research genuinely shows about the link (including the trials that failed), and what to actually do about it.
Section 1 · The Foundation
Why the gut has a say in how your brain works
The gut and the brain are in constant, two-way conversation. It runs in both directions at once — bottom-up, from gut to brain, and top-down, from brain to gut — and it travels along two distinct kinds of road.
The first is neural. The vagus nerve is the fast, direct line, and the enteric nervous system — the gut's own network of several hundred million neurons — does a great deal of local processing before anything is sent upstream. The second is systemic: hormones, neurotransmitters, cytokines and microbial metabolites carried in the bloodstream. The neural line is quick and specific. The systemic line is slower, broader, and — for our purposes — far more modifiable, because what the microbiome pours into the bloodstream depends on what you feed it.
Dysbiosis is not an infection, and it is not a single missing bug. It describes a community whose output has shifted — fewer of the microbes that produce protective, barrier-fortifying compounds, and more of those producing inflammatory or barrier-degrading ones. Two people can have very different bacterial species lists and the same functional problem. That is why modern microbiome analysis reports what the community can make, not just who lives there.
Six channels, one destination
The gut influences the brain through at least six mechanisms. They are worth knowing individually, because each one leaves a trace you can measure or a lever you can pull.
Immune & inflammatory
Microbes and endotoxin activate gut immune cells. The resulting cytokines cross a compromised barrier, disturb the blood–brain barrier and prime microglia, the brain's resident immune cells.
Short-chain fatty acids
Fibre-derived butyrate and propionate fuel the gut lining, tighten both barriers, act as epigenetic signals through HDAC inhibition, and calm microglia.
Neurotransmitters
Gut microbes make and modulate GABA, serotonin and dopamine-related metabolites that shape mood, stress reactivity and cognition.
Tryptophan routing
One amino acid, three fates: serotonin, the kynurenine pathway, or neuroprotective indoles such as IPA. Which road tryptophan takes is partly a microbial decision.
Enteroendocrine & hormonal
Short-chain fatty acids trigger serotonin, PYY and GLP-1 release, and the microbiome helps calibrate the HPA axis — the body's central stress system.
Vagus & enteric nerves
The fast lane. Signals run bottom-up from gut to brain and top-down from brain to gut, which is why stress changes digestion and digestion changes mood.
Six mechanisms sounds like six problems. In practice they converge on one place. Whether the signal is a cytokine, an endotoxin fragment or a short-chain fatty acid, it has to get from the gut lumen into the circulation and then from the circulation into the brain. Everything runs through the barriers.
The two gatekeepers
There are two of them, and they are more alike than most people realise.
The intestinal barrier is a single layer of columnar epithelial cells covered by a MUC2 mucus layer, secretory IgA and antimicrobial peptides, held shut by apical tight junctions — claudin, occludin and ZO-1. Its job is to absorb nutrients and water while keeping microbes, dietary antigens and endotoxin out of the bloodstream. It is opened by zonulin signalling, by inflammatory cytokines such as TNF-α and IFN-γ, by LPS–TLR4 signalling, and — importantly — by simple fuel starvation of the colonocytes that maintain it.
The blood–brain barrier is built from specialised brain microvascular endothelial cells within the neurovascular unit, sealed by tight junctions dominated by claudin-5. It is considerably tighter than any peripheral vessel, and it is opened by exactly the things a compromised gut sends its way: circulating LPS and the cytokines TNF-α, IL-6 and IL-1β, which cause loss and mislocalisation of claudin-5, occludin and ZO-1 (Braniste 2014; Sweeney 2019).
Microbiome
Fermentation and tryptophan metabolism generate the signals — short-chain fatty acids, indoles including IPA, and lipopolysaccharide.
Gut barrier — gatekeeper one
Epithelium, tight junctions and mucus decide what enters the blood. Under-fuelled or inflamed, it lets more through.
Circulation
Metabolites, cytokines and endotoxin are distributed systemically. This is where "metabolic endotoxaemia" becomes measurable.
Blood–brain barrier — gatekeeper two
Endothelial tight junctions protect the central nervous system. The same inflammatory signals that opened gatekeeper one now act here.
Brain
Microglia, neurons and synapses — the functional target. Microglial priming is where gut-derived inflammation becomes neuroinflammation.
The elegant part — and the clinically useful part — is that the same molecule seals both. In germ-free mice, animals raised with no microbiome at all, the blood–brain barrier is leaky, with reduced occludin and claudin-5. Colonising those mice with a single butyrate-producing bacterium, or simply giving them sodium butyrate, restored barrier integrity (Braniste 2014). Propionate does something similar to human brain endothelial cells, protecting them from LPS-induced permeability (Hoyles 2018).
Three signals worth knowing by name
Of all the compounds crossing this system, three do most of the explanatory work. If you take nothing else from this article, take these.
| Signal | Direction | What it does at the barriers | Clinical handle |
|---|---|---|---|
| Butyrate | Protective | Fuels colonocytes and drives tight-junction assembly and mucus production in the gut; restores blood–brain barrier integrity, matures and calms microglia, and raises BDNF. | Highly modifiable. Fibre diversity, prebiotics and some probiotics raise production capacity. |
| Indole-3-propionic acid (IPA) | Protective | A PXR ligand that raises tight-junction proteins and lowers enterocyte TNF-α; crosses the blood–brain barrier, scavenges free radicals and restrains amyloid aggregation. | Raised by dietary fibre and polyphenols; measurable in serum. |
| Hexa-acylated LPS | Harmful | Breaches the epithelium and, via TLR4, loosens tight junctions — worsening its own translocation. In blood it disrupts the blood–brain barrier and activates microglia. | An Enterobacteriaceae or Proteobacteria bloom flags the risk. Protect the barrier to limit translocation. |
Colonocytes consume the large majority of the butyrate produced in the colon — it is their preferred fuel. That means a faecal short-chain fatty acid measurement tells you what was left over, not what was made. A low faecal butyrate can reflect poor production or excellent absorption, and a normal result is not reassurance. Shotgun metagenomics sidesteps the problem by measuring the genes for butyrate synthesis directly, giving you production capacity (Vital 2017; Louis & Flint 2017).
Fun fact: humans have no butyrate synthesis pathway of our own. Not a partial one, not an inefficient one — none. Every molecule of butyrate in your colon was made by a bacterium fermenting fibre you could not digest yourself. The cells lining your large intestine run primarily on a fuel that only your microbes can manufacture. That is not a nutritional footnote; it is an argument for taking fibre diversity seriously.
What happens when endotoxin gets through
Lipopolysaccharide is the outer-membrane component of Gram-negative bacteria, and it is the single most potent activator of innate immunity we know of. But not all LPS is equal. Its inflammatory potency depends on the acylation pattern of its lipid A core: hexa-acylated LPS is a full TLR4 agonist, while penta- and tetra-acylated forms are weak agonists or outright antagonists (d'Hennezel 2017). This is why total endotoxin load is a poor marker and the acylation pattern is the informative one.
When hexa-acylated LPS reaches the circulation, it binds TLR4/MD-2 on microglia and brain endothelium, activating NF-κB and driving IL-1β, TNF-α, IL-6 and the NLRP3 inflammasome. Two things then happen in parallel. Those cytokines degrade endothelial tight junctions, raising blood–brain barrier permeability and admitting still more endotoxin — a self-amplifying loop. And they induce indoleamine 2,3-dioxygenase, which diverts tryptophan away from serotonin and down the kynurenine pathway toward quinolinic acid, an NMDA agonist that drives excitotoxicity (O'Connor 2009).
How confident can we be that this matters for how someone feels? Reasonably. Give healthy volunteers a tiny intravenous dose of endotoxin — 0.8 nanograms per kilogram, an amount with no clinical illness attached to it — and within hours their IL-6 and TNF-α rise sharply, their mood drops and their state anxiety climbs, in a dose-dependent fashion (Reichenberg 2001; Grigoleit 2011). In animals, blocking IDO abolishes the depressive-like behaviour entirely, which tells you the mood effect runs through that pathway rather than alongside it (O'Connor 2009).
That last pair of numbers deserves a moment. In post-mortem Alzheimer's tissue, microbiome-derived LPS is not merely present but dramatically enriched, and it is found inside amyloid plaques, around blood vessels and encircling neurons (Zhao 2017; Zhan 2018). Because amyloid-β is itself a TLR4 agonist, endotoxin and amyloid set off the same alarm receptor and each keeps switching the other back on. Once that loop starts, it is self-sustaining.
I want to be careful here, because this is where gut–brain writing usually overreaches. Finding LPS in the Alzheimer's brain does not prove that gut endotoxin caused Alzheimer's disease. It establishes that a microbiome-derived molecule reaches the brain in disease, in quantity, and sits in exactly the place the pathology is worst. That is a strong reason to take the gut barrier seriously. It is not the same as proof of causation, and the next section is where we test how far the evidence really goes.
Section 2 · The Evidence
What the research actually shows about the gut–brain link
There are three tiers of evidence in this field, and conflating them is the commonest mistake made in both directions — by enthusiasts who treat association as proof, and by sceptics who dismiss a genuinely strong causal literature because the human trials are patchy. The tiers are: what the population data show, what happens when you transfer a microbiome into an animal, and what happens when you intervene in a person.
Tier one: the pattern in the population data
Start with depression, where the datasets are largest. A meta-analysis in JAMA Psychiatry pooled 59 case-control studies across depression, bipolar disorder, psychosis and anxiety, and found the same thing in each: depletion of Faecalibacterium and Coprococcus, two of the principal butyrate producers (Nikolova 2021). An independent synthesis in Molecular Psychiatry, covering 4,917 cases and controls across 44 studies, reached the same conclusion by a different route (McGuinness 2022).
The single most instructive study is the Flemish population cohort published in Nature Microbiology. In 1,054 people, with validation in a second cohort taking the total to around 2,100, Faecalibacterium and Coprococcus tracked consistently with higher quality-of-life indicators, while Coprococcus and Dialister were depleted in depression — and critically, that depletion survived correction for antidepressant use (Valles-Colomer 2019). That adjustment matters more than it sounds. Antidepressants, antipsychotics, proton pump inhibitors, laxatives and metformin all reshape the microbiome, and a great many published "disease signatures" are partly drug signatures. The Dutch Lifelines analysis of 7,656 people, which adjusted rigorously for psychotropic medication, found the butyrate-producer depletion held up (Brushett 2023).
The same function-level signal recurs elsewhere. In Parkinson's disease, a species-level shotgun meta-analysis found Roseburia intestinalis and Faecalibacterium prausnitzii decreased, robustly and after adjustment for confounders, alongside increased Akkermansia muciniphila (Nishiwaki 2024). A large metagenomic study of 490 people with Parkinson's against 234 controls found over 30% of gut microbes, genes and pathways altered — short-chain fatty acid producers down, immunogenic LPS and curli biosynthesis up (Wallen 2022). In ME/CFS, a multi-omic study of 106 cases against 91 controls found reduced F. prausnitzii and Eubacterium rectale, with both metagenomic and metabolomic confirmation of deficient butyrate synthesis capacity — and F. prausnitzii abundance tracked inversely with fatigue severity (Guo 2023).
On the other side of the ledger, the endotoxin signal shows up too. In Parkinson's, 11 of the 55 disease-enriched species are Gram-negative, several carrying highly stimulatory LPS, with lipid A biosynthesis increased overall (Wallen 2022). In multiple sclerosis, a metagenome-wide association study found "bacterial cell outer membrane" the top-enriched functional term and the LPS biosynthesis pathway significantly altered — though in a small cohort of 26 against 77 (Kishikawa 2020). In cognitive impairment, amyloid-positive patients carry more Escherichia/Shigella, and that abundance correlates with circulating IL-1β, NLRP3 and CXCL2 (Cattaneo 2017).
Depression is the honest weak spot here. The endotoxin link in mood disorders still rests largely on serology — raised IgA and IgM against the LPS of Gram-negative enterobacteria in chronic depression — rather than on metagenomics (Maes 2012). It is the thinnest of the four, and worth saying so.
Tier two: transfer the microbiome, transfer the phenotype
Association studies cannot tell you which way the arrow points. Faecal transplant into germ-free or microbiota-depleted rodents can. The design is simple: take the microbiome from a person with a condition, put it into an animal with no microbiome of its own, and see what happens to the animal.
| Condition | What transferring the human microbiome did | The detail that matters |
|---|---|---|
| Parkinson's disease | Microbiota from people with Parkinson's worsened motor deficits in α-synuclein-overexpressing mice, compared with healthy-donor microbiota (Sampson 2016). | The microbiome was required for the deficits: antibiotics rescued them and recolonisation restored them. Short-chain fatty acids alone drove neuroinflammation. |
| Autism | Human donor microbiota from autistic individuals produced hallmark autism-like behaviours in the offspring of germ-free mice (Sharon 2019). | Two candidate metabolites — 5-aminovaleric acid and taurine — shifted both behaviour and neuronal excitability. |
| Multiple sclerosis | MS-patient microbiota worsened autoimmune neuroinflammation, demonstrated by two independent laboratories (Cekanaviciute 2017; Berer 2017). | T cells skewed pro-inflammatory (Th17 up, Treg down). Across 34 MS-discordant identical twin pairs, patient microbiota raised spontaneous disease. |
| Depression | Microbiota from depressed patients transferred depression- and anxiety-like behaviour to microbiota-depleted rodents (Kelly 2016; Zheng 2016). | Again two independent labs. Recipients showed anhedonia and behavioural despair, with altered host carbohydrate and amino-acid metabolism. |
This is the strongest causal evidence the field has: four brain conditions, multiple independent laboratories, and in each case the phenotype travels with the microbiome. It is where the link stops being a correlation. What it is not is proof of a treatment — establishing that a microbiome can cause a phenotype in an animal is a different claim from establishing that changing it fixes the phenotype in a person, and the second question has a messier answer.
Tier three: what happens when you intervene in humans
Faecal microbiota transplantation in people is the sharpest test available, and the results form a ladder rather than a verdict.
At the top sits autism, where an intensive open-label microbiota transfer protocol produced roughly a 45% reduction in core symptom scores that was still present at two-year follow-up — the largest and most durable signal in the literature (Kang 2019). It is open-label, which is a real limitation in a condition where expectation effects are substantial, but the durability is difficult to dismiss.
Parkinson's disease has more randomised trials than anything else here. The GUT-PARFECT phase 2 trial randomised 46 people with mild-to-moderate disease to healthy-donor or autologous transplant. At twelve months the donor group improved by an average of 5.8 points on the MDS-UPDRS motor scale against 2.7 points for placebo, with better colon transit times and no serious adverse events (Bruggeman 2024). Constipation and non-motor benefit are the most reliable findings across trials; motor effects remain mixed.
Below that, the ladder gets thinner. In Alzheimer's disease and cognitive decline, small case series show cognitive gains — but so far only in patients being treated for Clostridioides difficile infection, which is a very particular situation. In anxiety, improvement appears alongside gut symptom improvement in IBS trials, not as a standalone effect. Bipolar disorder has one striking positive case report and a trial now underway.
And at the bottom sit the results nobody quotes. In multiple sclerosis, transplant is safe but efficacy is unproven — the pilot ended early. In ME/CFS, a blinded pilot showed no symptom benefit at all. In schizophrenia, there are no human trials whatsoever; the causal signal exists only in mice.
"Mechanism-plausible causation" is the honest description of where this evidence sits, and it is worth saying out loud to patients. The animal causal data are genuinely strong. The human treatment data are strongest in Parkinson's and autism, mixed in mood disorders, and negative in ME/CFS. Presenting the gut–brain axis as settled therapeutics erodes trust when a patient later reads the negative trials. Presenting it accurately — a coherent, actionable model with real gaps — holds up.
What about probiotics and prebiotics?
This is the question patients ask, and the honest answer is more encouraging than the faecal transplant literature. A 2025 systematic review and meta-analysis in Nutrition Reviews, restricted to clinically diagnosed samples rather than healthy volunteers, pooled 23 randomised controlled trials in 1,401 patients. Probiotics produced a substantial reduction in depression symptoms across 18 trials (SMD −0.96, 95% CI −1.31 to −0.61) and a moderate reduction in anxiety across 9 trials (SMD −0.59, 95% CI −0.98 to −0.19). Prebiotics, on only three trials and 144 participants, showed a non-significant trend for depression (Asad 2025).
Two caveats keep that in proportion. Heterogeneity was high, and roughly a third of the included studies carried a high risk of bias — so the true effect is probably smaller than the headline number. And the effects are strain-specific: the trials that worked used particular organisms at particular doses, and the results do not transfer to whatever is on the shelf. A probiotic is not a category, it is a product.
Four caveats that change how you read any of this
Confounding is the rule, not the exception
Diet dominates. Much of the autism microbiome "signature" is explained by restricted eating patterns rather than the condition itself (Morton 2023). Medications and gut transit reshape the community independently. The strongest studies adjust for all of this; associations that survive adjustment are the trustworthy ones.
Almost all human data are cross-sectional
Directionality exists only where transfer or colonisation has actually been tested. Everywhere else, present the link as mechanism-plausible correlation.
Heterogeneity is substantial
Effect directions shift with geography, diet and bioinformatic pipeline — Bacteroides even reverses between US and Chinese Alzheimer's cohorts. Trust uniform-pipeline meta-analyses over any single cohort, and treat very high classifier accuracies from small samples as prone to overfitting.
Taxa are context-dependent
There are no simple good bugs and bad bugs. Species within Bacteroides and Ruminococcus pull in opposite directions on the same biomarkers — which is precisely the argument for working at species level rather than genus.
Fun fact: Akkermansia muciniphila is one of the most-hyped organisms in the microbiome world — and one of the best illustrations of why hype is unwise. It is consistently enriched in Parkinson's disease, where it is thought to reflect mucin degradation, yet it appears protective in studies of cognition and metabolic health. Same organism, opposite implications, depending on the context it sits in. Any report that grades bacteria as simply good or bad is telling you something the biology does not support.
Section 3 · The Clinical Work
Treatment: what to remove, what to restore, and how strong the evidence is
If everything above converges on barrier integrity, then the clinical work is barrier work. And barrier work has a natural order: remove first, restore second. Removing what is actively damaging the gut lining costs nothing, takes effect quickly, and stops you pouring repair agents into a system that is still being broken. Restoration on top of ongoing damage is expensive and slow.
Everything in this section is graded. Where there is human brain-outcome evidence, I say so. Where the evidence is barrier data with a mechanistic brain rationale, I say that instead. The distinction is not academic — it determines what you can honestly promise.
Remove: what is driving intestinal permeability
| Driver | Evidence at the gut barrier | Brain relevance |
|---|---|---|
| NSAIDs Strong · Human RCT | Five days of indomethacin raised small-bowel permeability roughly three-fold in healthy volunteers (lactulose:rhamnose 0.35 → 0.88) (Mahmood 2007). | Relevant through the endotoxin route rather than any direct neurological effect. The barrier breach is the brain-facing problem. |
| Alcohol Strong · Human | A single binge raises serum endotoxin and bacterial DNA in healthy adults; chronic intake impairs barrier function (Bala 2014). | Alcohol use disorders were a major risk factor for dementia of all types — especially early-onset — in a nationwide cohort of over one million people (Schwarzinger 2018). |
| Chronic psychological stress Strong · Mechanism | Public-speaking stress raised intestinal permeability — but only in cortisol responders. A mast cell stabiliser blocked it; CRH reproduced it (Vanuytsel 2014). | The same CRH–mast cell axis that opens the barrier sits underneath stress-related mood and gut symptoms. One mechanism, two outcomes. |
| Low-fibre, ultra-processed pattern Strong · Cohort | Fibre deprivation drives the microbiota to degrade the colonic mucus layer itself, shown definitively in gnotobiotic mice (Desai 2016). | Highest versus lowest ultra-processed intake: roughly 50% higher depression risk in 31,712 people, and faster cognitive decline (Samuthpongtorn 2023; Gomes Gonçalves 2023). |
| Excessive exercise load and heat Moderate · Nuanced | Heavy, prolonged or heat-stressed exercise transiently raises permeability in a dose-dependent way (Davison 2016). | Do not remove exercise. Physical inactivity is itself a modifiable dementia risk factor. Reduce dose, heat and duration instead (Livingston 2024). |
| Dietary emulsifiers Emerging | Carboxymethylcellulose reduced microbial diversity and drove microbiota encroachment toward the epithelium in a controlled-feeding randomised trial (Chassaing 2022). | Behavioural effects are preclinical only so far. Treat as a plausible additive contributor, not an established brain risk. |
Three of those six — alcohol, ultra-processed food and physical inactivity — have independent human evidence for worse brain outcomes, not merely worse barriers. Addressing them is the cheapest intervention available and the one most likely to move both endpoints at once.
Patients who read that exercise raises intestinal permeability sometimes stop exercising. That trade is clearly unfavourable. The permeability rise is transient and load-dependent — it is a feature of prolonged, high-intensity or heat-stressed sessions, not of the walking, resistance training and moderate cardiovascular work that carry the cognitive benefit. In the endurance athlete with gut symptoms, adjust session duration, ambient heat and pre-exercise NSAID use before touching training volume, and consider zinc carnosine with bovine colostrum, which truncated the exercise-induced rise by around 70% in a controlled trial (Davison 2016).
Restore: agents with actual human permeability data
Here is where honesty about evidence tiers earns its keep. Of the interventions commonly used for gut barrier repair, only two have direct human brain-outcome evidence. The others are barrier agents with a mechanistic brain rationale. Both categories have a place. They should not be described the same way.
| Intervention | Studied dose | Gut barrier evidence | Brain evidence |
|---|---|---|---|
| Dietary fibre → SCFA | Food first; diversity over quantity | Raising intestinal short-chain fatty acid concentration improves gut permeability across studies in systematic review (Pohl 2022). | Direct Butyrate matures microglia and tightens the blood–brain barrier; dietary pattern data support cognition (Braniste 2014; Morris 2015). |
| Zinc repletion | Correct measured deficiency | In patients with quiescent Crohn's and elevated permeability, oral zinc improved intestinal permeability (Sturniolo 2001). | Direct Peripheral zinc is lower in depression across 17 studies and 1,643 cases; adjunctive zinc lowers symptom scores (Swardfager 2013; Lai 2012). |
| Zinc carnosine | 37.5 mg twice daily | Prevented the three-fold indomethacin-induced permeability rise; cut exercise-induced permeability by around 71% (Mahmood 2007; Davison 2016). | Indirect No brain data for the carnosine complex itself. The brain case rests on zinc status. |
| L-glutamine | 5 g three times daily × 8 weeks | Normalised the lactulose/mannitol ratio (0.11 → 0.05) in post-infectious IBS-D — though a 10-study meta-analysis found no overall effect (Zhou 2019; Abbasi 2024). | None No human brain trials. Use caution where ammonia handling is impaired. |
| Saccharomyces boulardii | 200 mg three times daily × 3 months | Improved lactulose/mannitol in Crohn's remission (n = 34) while placebo worsened; a larger relapse trial was negative (Garcia Vilela 2008; Bourreille 2013). | None Strain-specific: the mood evidence sits with other probiotic strains, not this one. |
| Bovine colostrum | 20 g/day × 14 days | Truncated the rise in permeability after heavy exercise, additively with zinc carnosine (Davison 2016). | None Include for barrier repair, not for a neurological claim. |
Diet is the one lever that moves everything at once
No supplement on that table moves as many markers simultaneously as the plate does. In the NU-AGE trial, a one-year Mediterranean dietary intervention in older adults across five European countries shifted the microbiome toward short-chain fatty acid producers, raised diversity, reduced frailty and improved health status — while the taxa associated with inflammation moved in the opposite direction (Ghosh 2020). One intervention, every core marker.
Fermented foods add something distinct. In a 10-week randomised trial at Stanford comparing a high-fermented-food diet against a high-fibre diet in 36 healthy adults, the fermented food arm increased overall microbial diversity and lowered 19 inflammatory proteins including IL-6. The high-fibre arm held diversity steady and did not lower those markers over the same period — which is not an argument against fibre, but an argument that the two do different jobs and belong together (Wastyk 2021).
On the MIND diet, the story is instructive and I would rather you heard the full version. The observational data were striking: adherence was associated with meaningfully slower cognitive decline with ageing (Morris 2015). Then it was tested properly. A three-year randomised trial in 604 older adults published in the New England Journal of Medicine found that both groups improved their global cognition scores, with no significant difference between the MIND arm and the control arm (Barnes 2023). That does not make the MIND diet a bad idea — both arms were mildly calorie-restricted and both improved — but it does mean the confident claim that a specific dietary pattern prevents cognitive decline is not supported by randomised evidence. Recommend the pattern for its cardiometabolic, microbiome and general-health benefits, which are well established. Do not promise it prevents dementia.
Prebiotics: the doses that were actually studied
Prebiotic dosing in practice is frequently guesswork. It does not need to be — most of these have a studied dose, and the gap between the studied dose and the label dose is often the reason a trial of a prebiotic "did not work".
| Prebiotic | Studied dose | Best-supported effect | Note |
|---|---|---|---|
| GOS / B-GOS | 5.5 g/day | Lowered the waking cortisol response and shifted emotional attentional bias (Schmidt 2015; Vulevic 2008) | The strongest gut–brain data of any prebiotic. FOS had no effect in the same trial. |
| Partially hydrolysed guar gum | ~5–6 g/day | Butyrogenic; well tolerated and low-gas | The tolerable option for sensitive guts. No cognitive RCTs. |
| Resistant starch | ~15–40 g/day | Strongly butyrogenic | Excellent for butyrate capacity. No direct cognitive RCTs. |
| Beta-glucan (oat) | ~3 g/day | Fermentable; supports SCFA production | This is the cardiometabolic dose; cognitive benefit is indirect. |
| Inulin / FOS | ~5 g/day, up to 10–20 | Bifidogenic | Reserve for microbiome shift and laxation rather than mood. |
| Pectin | ~10–15 g/day | Fermentable; supports SCFA and barrier function | Cognition indirect. |
| HMOs (e.g. 2'-FL) | ~1–5 g/day | Bifidogenic | Emerging adult data; no cognitive RCTs yet. |
Start low and increase slowly — a quarter of the target dose for the first week is rarely too cautious. In patients with existing bloating, visceral hypersensitivity or a history of small intestinal bacterial overgrowth, begin with partially hydrolysed guar gum rather than inulin, which is the most gas-forming of the group. If you want the stress-physiology effect specifically, B-GOS at 5.5 g/day is the only prebiotic with dedicated human data behind it, and it needs three weeks before reassessing. And check the label maths: many products deliver a fraction of the studied dose per serve.
The second gatekeeper: what the blood–brain barrier evidence supports
The blood–brain barrier is a harder clinical target, for a simple reason: you cannot measure it in routine practice. Assessing it properly requires dynamic contrast-enhanced MRI or cerebrospinal fluid markers. So the work here is driver removal rather than monitoring — and, encouragingly, three of the main drivers are the same levers you have already pulled at the gut wall.
Blood pressure control
The strongest lever on this list. Intensive versus standard systolic control in 9,361 people cut mild cognitive impairment by 19% and slowed white matter lesion progression — though probable dementia alone did not reach significance (Williamson 2019; Nasrallah 2019).
Glycaemic and metabolic control
Chronic hyperglycaemia injures cerebral endothelium and pericytes, and diabetes is an established modifiable dementia risk factor. Shares nearly every dietary lever with gut barrier repair (Livingston 2024).
Physical activity
Improves cerebrovascular function and perfusion. There is no human blood–brain barrier permeability trial — recommend it for the cognitive evidence, which is strong, not for a barrier claim (Livingston 2024).
Alcohol reduction
One of very few levers with strong evidence at both barriers and for dementia risk. Raises circulating endotoxin acutely and is directly neurotoxic; the two routes compound (Bala 2014; Schwarzinger 2018).
Butyrate capacity
The strongest shared lever between the two barriers. The same fibre-driven capacity that fuels the gut lining tightens the blood–brain barrier and matures microglia — in animals (Braniste 2014; Erny 2015).
Tryptophan indoles
IPA crosses the blood–brain barrier and signals through PXR on brain endothelium, raising tight-junction proteins and dampening TLR4 inflammation. Human data remain associative (Venkatesh 2014; Owe-Larsson 2025).
No nutritional or microbiome intervention has yet been shown to improve measured blood–brain barrier permeability in a human being. Blood pressure control is the only lever here with randomised human evidence on a brain endpoint. The two-barrier model is mechanistically coherent and the gut-side evidence is genuinely human — but the chain from "repaired the gut barrier" to "improved the blood–brain barrier" to "better cognition" has never been demonstrated end to end in people. Present it as the working model that organises the markers, not as a proven therapeutic sequence.
That gap changes how you build a plan. You are not treating a proven causal chain; you are addressing a set of drivers that each carry their own evidence — some for the gut, some for the brain, a few for both — and that happen to converge on the same interventions. Convergence is exactly what you want clinically. The person who reduces alcohol, eats thirty different plants a week, manages their blood pressure, moves daily and handles their stress load is improving their gut barrier, their microbial output, their cerebrovascular health and their dementia risk profile at the same time. You do not need the chain to be proven for the actions to be right.
A clinical workflow
Assess
Identify the phenotype. Take a full dietary, antibiotic, metabolic and gastrointestinal history. Order stool metagenomics where it will change the plan, and consider metabolite testing alongside.
Interpret
Read the balance rather than the bug list — butyrate capacity, endotoxin and Proteobacteria load, tryptophan routing — and always as a pattern alongside the clinical picture.
Target
Remove the drivers first. Then diet-first restoration: fibre diversity, a Mediterranean pattern, fermented foods, and reduced ultra-processed intake. Add strain-specific probiotics and targeted barrier agents where indicated.
Re-measure
Track symptoms and relevant inflammatory markers, then re-test to confirm the microbiome actually shifted. Microbiome data informs clinical judgement; it does not replace it.
Miriam, 52 — brain fog, low mood and twenty years of an irritable gut
A fictional, de-identified composite created for illustration. It is not a real patient and is not a treatment protocol.
Presentation. Eighteen months of worsening brain fog and word-finding difficulty, flat mood, poor stress tolerance and waking unrefreshed. Twenty-year history of mixed-type IBS. Three courses of antibiotics in two years. Daily ibuprofen for lower back pain. Four to six standard drinks weekly, concentrated on two nights. Shift work, with an estimated fibre intake of 14 g/day and around eight different plant foods per week. Blood pressure 148/92 on two readings. Not taking antidepressants.
Metagenomic profile. Low butyrate synthesis capacity, low IPA-producing capacity, elevated hydrogen sulphide producers and mucin degraders, modest Proteobacteria enrichment, diversity in the low-normal range.
Phase one — twelve weeks
- Daily NSAID stopped, with GP review; back pain managed with topical therapy and physiotherapy
- Alcohol limited to two non-consecutive occasions per week
- Plant diversity target of 30 per week, tracked; started at 8
- Fibre raised from 14 g to 30 g/day over six weeks, with PHGG 3 g/day increasing to 6 g
- Two serves of fermented food daily, introduced from week two
- Ultra-processed intake reduced by reworking the two shift meals
- Zinc repleted after a low plasma zinc was confirmed
- Blood pressure referred back to GP and management commenced
- Ten minutes of daily stress downregulation, treated as non-negotiable, plus an evening email boundary
Reassessment — sixteen weeks
- Brain fog markedly reduced; word-finding difficulty resolved
- DASS-21 stress subscale moved from severe to mild
- Bristol stool type 6–7 settled to type 4; bloating substantially reduced
- Plant diversity sustained at 28–32 per week
- Repeat metagenomics: butyrate synthesis capacity into the reference range, hydrogen sulphide producers reduced, diversity improved
- Blood pressure 128/82 under GP management
- Still present: a mild afternoon energy dip
Note the honest limitation. Nine things changed at once, so no single intervention can be credited — and that is both normal and appropriate. The gut–brain axis is not a single-lever system, and treating it as one is how people end up disappointed by a supplement that was never going to work in isolation.
In Summary
Six principles to take away
Read function, not a bug list
Across nine conditions the taxa disagree but the function repeats. Work at species level and interpret what the community can make, not who is present.
Seal the barrier first
Immune signalling, SCFAs, neurotransmitters, tryptophan routing, gut hormones and the vagus all converge on barrier integrity. Every other lever runs through it.
Call it mechanism-plausible causation
Transferring patient microbiota reproduces the phenotype in animals. In humans the signal is strongest in Parkinson's and autism, and negative in ME/CFS. Never call it a cure.
Know the three signals
Butyrate and IPA up, hexa-acylated LPS down. Almost every intervention in this article is an attempt to move those three in the right direction.
Remove before you restore
Alcohol, NSAIDs, chronic stress and ultra-processed food all raise permeability in humans. Removing them is free and works faster than anything you can prescribe on top.
Name the evidence tier every time
Zinc and dietary fibre are the only agents here with human brain-outcome data. Blood pressure control is the only randomised brain endpoint. Say which is which.
The gut–brain axis has been oversold and undersold in roughly equal measure. Oversold as a cure for conditions it will not cure; undersold by clinicians who dismissed it before the mechanistic work matured. The truthful position sits between the two, and it is more useful than either. There is a reproducible functional signature across brain conditions. There is a coherent set of mechanisms connecting it to neuroinflammation. There is genuine causal evidence in animals, mixed evidence in human trials, and a set of interventions — fibre diversity, fermented foods, alcohol reduction, stress load, NSAID use, blood pressure, movement — that are worth doing on their own merits regardless of how the remaining questions resolve.
That is not a hedge. It is the strongest honest case, and it happens to point at the same actions the enthusiastic version does — just with an accurate account of why.
Further Your Practitioner Education — Live Event
Are you a practitioner who wants to take this further? Everything above comes from a much bigger clinical conversation and Dr Brad Leech is unpacking all of it at ACNEM's The Neuro Blueprint: Integrative Approaches to Cognitive Health, where he presents From Dysbiosis to Brain Function. The session goes deeper into how gut dysbiosis drives neuroimmune activation and blood-brain barrier disruption, the mechanisms carrying that signal cytokine signalling, endotoxin translocation and microbial neurotransmitter production — the specific microbiome patterns that tell you whether the gut is genuinely a driving factor in a patient's brain function, and clear treatment objectives for restoring both intestinal barrier integrity and the blood-brain barrier. Join a full day dedicated to brain health on Saturday 17 October 2026 at UTS Sydney or live-streamed online — secure your place at ACNEM.
Dr Brad Leech
Brad is a PhD-qualified Clinical Nutritionist and Herbalist specialising in chronic autoimmune conditions and complex gastrointestinal disorders. He provides complete and personalised care to his patients using functional nutrition, integrative medicine and holistic wellness.