Why taurine runs low in the spinal fluid in Parkinson’s

Taurine is a β-amino sulfonic acid, meaning it carries a sulfonic acid group where an ordinary amino acid has a carboxyl group, and that its amino group sits on the β-carbon. That is why it will not fit into a peptide chain. Adults make it themselves from cysteine, via cysteine dioxygenase and cysteine sulfinic acid decarboxylase, on the order of 50 to 125 mg a day against a dietary intake of 40 to 400 — enough to call it conditionally essential.^2,3 Its main known functions are bile acid conjugation, osmoregulation, membrane stabilisation and modulation of calcium flux.

Taurine is also a weak agonist at GABA_A and glycine receptors, and in the lab it opens the same chloride channels as glycine.¹³ Häusser, Yung and Lacey saw in 1992 that dopamine neurons from the rat substantia nigra stopped firing spontaneously because of it, and strychnine — the poison that shuts precisely those channels — abolished the effect entirely. The lowest they put in the bath for that was three hundred micromoles per litre, and they did not try lower.¹⁴ Spinal fluid holds a fraction of that: in people who took nine grams a day for a year it went from 11 to 42 micromoles per litre.¹⁵ And what actually reaches the receptor in such a slice is lower still than what is in the bath, because the glial cells draw it off — they wrote that themselves.

What taurine does inside a cell is something else again, and that is PART III. And both are the opposite direction from the widely repeated claim that taurine “raises dopamine”; why that one does not hold is in PART VI.

This concerns 45 drug-naïve patients with a mean age of 68.6, in whom amino acids were measured in both plasma and cerebrospinal fluid, alongside dopamine transporter imaging. Higher CSF taurine went with better preserved DAT binding and with better motor and daily-living scores; in multivariate analysis CSF taurine remained an independent factor.¹⁰

That fits the wider picture. A meta-analysis of amino acids as biomarkers found taurine lower in cerebrospinal fluid in Parkinson’s, but no significant difference in serum or plasma.¹⁶ The signal sits in the compartment that matters, not in the tube of blood taken at the surgery.

In 2026 a broad metabolomics study was added to that, with two independent cohorts: 153 drug-naïve de novo patients against 153 matched controls, and 135 against 135. Taurine was among the metabolites contributing most stably to the separation and stayed selected when the analysis was restricted to drug-naïve participants only.¹⁷ As far as I can tell the accessible main text does not give the direction: whether plasma taurine in these patients was higher or lower is not in it. What it does show is taurine standing out again, in two groups at once, and not as a by-product of medication.

And here is why it matters that these people were not yet on medication. In other work plasma taurine in Parkinson’s patients was lower than in controls and tracked motor severity — but treated patients had lower levels than untreated ones, with a negative correlation to cumulative levodopa dose that survived adjustment.¹⁸ So part of what you measure in blood is a medication effect, not a disease effect.

What this design cannot do is separate cause from effect: a cross-sectional measurement shows that two things move together and not which one is driving, so low taurine could just as easily be a consequence of the neurodegeneration as a cause of it.

The sharpest counterweight comes from a primate study in which the researchers split their exposed animals into four groups rather than two. They gave one group of macaques MPTP, sorted those afterwards by how it had turned out — never fell ill, recovered, moderate and severe parkinsonism — placed alongside them animals that had never been given the toxin, and only then measured the taurine, in caudate, putamen, primary motor cortex and prefrontal cortex. Taurine was higher in the asymptomatic and recovered animals than in both the parkinsonian animals and the controls, most clearly in cortex. They were deliberately looking for brain constituents that could not themselves be a consequence of the degenerative process.¹⁹

Three limits go with it. No taurine was administered, so it stays a correlation, only then in a far better species and in tissue rather than in fluid. Taurine was not the only difference either, because MAO B was higher in the severely affected animals, and MAO B is precisely the enzyme that turns MPTP into the compound that kills the cells — so the authors end up concluding two things at once, less protective taurine and more bioactivating MAO B. And measurement was after the fact, so whether the taurine difference predated the exposure or is a response to it cannot be read off these data.

For the direction question it helps regardless: taurine high in tissue where there was no damage, in animals that did get the toxin. But the real answer to “does taurine do something, or does it merely happen alongside?” only comes when you take it away, and that is the next part.

Che picked paraquat and maneb, two pesticides used alongside each other on farms, and gave them to C57BL/6J mice twice a week for six weeks at 10 and 30 mg/kg. That is a much slower model than the acute injection these experiments usually run on, and presumably it was chosen for that reason, since someone exposed to pesticides in a field does not take them in all at once either. The taurine went in thirty minutes before each exposure, 150 mg/kg into the abdomen.²⁰

What they then counted is the endpoint that matters, namely the TH⁺ neurons in the substantia nigra pars compacta, blinded, on coded slides and in every third serial section. A formal stereological estimator it is not, but it is the measurement the weaker studies skip. At 2, 4 and 6 weeks they watched the loss climb, the animals given taurine kept significantly more of their neurons, and the gait analysis moved with it — stride length and distance covered.

High-molecular-weight α-synuclein bands also fell, though that is probably an outcome rather than a mechanism, since nobody tested whether taurine binds or inhibits α-synuclein directly. The parsimonious reading seems to me that less oxidative damage yields less clumping downstream.

What taurine switched off in those animals they measured as well. It inhibited NADPH oxidase, visible as less membrane translocation of p47phox, less gp91phox and less superoxide, and it inhibited the NF-κB route, visible as less phosphorylation of p65, IκBα and IKKα. The M1 markers iNOS, TNFα and IL-1β fell while the M2 markers stayed unchanged, which seems to indicate that taurine takes out the inflammatory setting without driving up the clean-up setting.

The design matters here, because the depletion experiment was not in the mice. They ran it in primary midbrain cultures from rat embryos (E14), from which the microglia were removed with leucine methyl ester — clearing more than 99.9%. In those cultures taurine’s protective effect on the TH⁺ neurons was afterwards completely gone.²⁰

That figure alone misleads without the other one. With microglia present, 49.4% of the TH⁺ neurons were lost, and taurine brought that down to 29.2% at 25 µM and 8.2% at 50 µM. Take the microglia out and the loss caused by the poison itself already fell back in part — the authors call it a partial attenuation and give no percentage for it, so the two conditions cannot be weighed against each other.

And there a question forces itself on you: if taurine works only through microglia, how can it prevent more loss than removing those same microglia does? Probably because a depletion experiment gives no upper bound at all. Microglia do two things at once — they cause damage and they clear up — and whoever removes them removes both. Taurine leaves them in place and takes out only the attack mode; that is precisely what Che measured above when the M1 markers fell while the M2 markers did not move. A cell you convert can therefore end up better than a cell that is gone. That comparison was not measured, as it would require laying both conditions side by side.

What follows regardless is the narrower fact this part rests on: without microglia, taurine does nothing. And the poison damaging neurons even without those cells means microglia amplify the damage rather than cause it — so taurine acts on that amplification, not on the poison.

So this is not a third arm of the same mouse experiment but a second system: different species, embryonic tissue, outside the body. A depletion in a living animal would have been stronger. What there is instead is an effect in a living animal plus a separate test of necessity pointing the same way — and as far as I can tell it is the only experiment showing a mechanism rather than a correlation.

Microglia are the best-studied glial cell here, but not the only one involved. In MPP⁺-exposed primary astrocytes and MPTP mice the picture fitted suppression of astrocytic mitochondrial stress and inflammatory signalling — astrocytes were not removed or blocked there, so “the protection ran through astrocytes” is not what it says.²¹ In the rotenone model taurine suppressed activation of microglia and astrocytes side by side.²² So “taurine acts on the inflammatory cells around the dopamine neuron” holds; “taurine works exclusively through microglia” is too narrow — though Che’s depletion experiment remains the only one that actually removes a route.

NF-κB recurs in Che (mouse microglia), in Moon (astrocytes) and in the rotenone work. And the node is not rodent-specific — for NOX2 the human bridge is the most direct. In postmortem substantia nigra from people with Parkinson’s, NOX2 was highly active, in dopamine neurons as well as microglia. In acute and sub-acute animal models only neuronal NOX2 was active, suggesting the dopamine neuron goes first and the glia follow — a self-reinforcing loop rather than “it starts in the microglia”.²³ Here too, no taurine was involved.

An adjacent human network has been mapped separately. In human neuronal and microglial cell lines and in iPSC-derived dopaminergic neurons and microglia, α-synuclein aggregates damage the mitochondria, after which mitochondrial DNA leaks into the cytosol and the cGAS–STING–NF-κB–IRF3 route is activated; that promotes tunnelling nanotubes and transfer of α-synuclein and damaged mitochondria between neurons and microglia.²⁴

The hard limit alongside it: no taurine was used in that study. It therefore proves nothing about taurine. And it is not the same switch as Che’s: Che measured NOX2 and the IKK–IκB–NF-κB axis, Chakraborty manipulated cGAS/STING. What both human studies together do show is that the inflammatory node taurine acts on in the mouse exists in human dopamine neurons and microglia. That is foundation, not roof.

One strong study is not proof. This is what else is out there, ordered by how hard the endpoint is:

The bottom three belong here precisely because they do not play along. In the rat study the behaviour improved while the fall in dopamine was not significantly reversed. The rotenone study says rats in its title and mice in its methods section — anyone taking both at their word has to assume the animals changed species somewhere between the abstract and the cage. It also finds no dose-response relationship, and looks not at the substantia nigra but at the prefrontal cortex.

And then Navneet, the one outright negative study in the file. Taurine did not reverse the fall in dopamine after MPTP, scavenged no hydroxyl radicals in isolated mitochondria, and without any toxin even produced a small drop in striatal dopamine.²⁹ That bounds any simple account in which taurine is a radical scavenger everywhere.

One difference is decisive though, and it sits in the methods: here the taurine started after the second MPTP injection, whereas Che and Moon dosed beforehand. So this is a treatment experiment, not a prevention one. It also ran six animals per group and measured only neurotransmitters at four days — no cell counts. That makes it no evidence that 850 mg orally is harmful; it is a reason not to read “taurine protects” as a general property.

And then the limit of all of it: these are toxin models. You poison a mouse, you see whether a compound damps it down, and that is not a model of the slow human disease — the history of neuroprotectants that worked here and failed in the clinic is long. They are not to be waved away, but not to be read as prediction either.

Against that stands one finding that does not come from a toxin model. In 2018 Graham and colleagues injected α-synuclein fibrils into the olfactory bulb of mice, after which the clumps spread through the brain on their own, and then went looking at which metabolic pathways in that brain were most disturbed. Taurine and hypotaurine metabolism came out on top.³⁰ No taurine was administered in that experiment, so it says nothing about treatment. What it does do is make taurine surface in a model where nobody poisoned the brain.

The tail is called 5-taurinomethyluridine, τm⁵U, and it sits at the wobble position — position 34 — of a handful of mitochondrial tRNAs. In 2018 Asano, Suzuki and colleagues worked out how it is made: out of taurine and 5,10-methylene-tetrahydrofolate, by the enzymes MTO1 and GTPBP3. Extracellular taurine is taken into the cell and there, in their own words, used as a direct metabolic substrate.⁵

They showed it is substrate-driven in three ways. HeLa cells in taurine-free medium lose τm⁵U and get it back with 40 mM taurine added. Cells without GTPBP3 have respiratory defects and make less mitochondrial protein, and in cells from a patient with a GTPBP3 mutation there was barely any τm⁵U to be found. And in animals that cannot make taurine themselves the modification moves with the diet.

That last one is smaller than it sounds, and they say so themselves. In a cat on a taurine-poor diet the modification went from 86 to 63 per cent, in Japanese flatfish from 46 to 35 — and for the cat that was one animal against one control, whose liver came from a cat hit by a car. What stands is that the modification moves with the diet. How steeply it runs in a mammal that does make its own taurine is not in there.

In taurine-poor cells something else turns up at that position: cmnm⁵U, the same basic form but with glycine instead of taurine. The authors propose it as a marker for taurine starvation and call it a backup system. How well that backup decodes is not known — they write that explicitly.

Then Kirino’s experiment, which measures the link directly. With what they call molecular surgery they cut the tRNA in two, stripped the taurine modification off the wobble base with periodate oxidation, and stuck it back together. What was left was a tRNA missing only the taurine modification, with no pathogenic mutation. That operated tRNA translated UUG severely worse while UUA did not suffer at all, and aminoacylation was equal in both cases — 88 per cent for the wild type, 87 for the operated one. So it is not a charging problem. By comparison, the MELAS mutations themselves did lower aminoacylation, to 48 and 80 per cent.³¹

The arithmetic is in that same paper. ND6 contains eight UUG codons, which is 42.1 per cent of all leucine codons in that gene. And in human cybrid cells with rising MELAS mutation load, translation of ND6 falls specifically and steeply, while total mitochondrial protein synthesis does not.

Schaffer and colleagues laid it out in 2013: twelve of the thirteen mitochondrially encoded proteins have more UUA than UUG codons, ND6 has eight UUG, and it follows that a taurine deficiency should hit only complex I hard. The other MELAS defect — reduced aminoacylation, which touches both UUA and UUG — should hit complexes I and III through V. Two defects, two different signatures, and the taurine one is the narrow one.³³

They do acknowledge a gap in it in 2013. At that point, they write, there was no experimental evidence that taurine deficiency actually lowers the modification; for that step they cited unpublished data. That is precisely the gap Asano and Suzuki closed five years later, with mass spectrometry, in cells and in animals, published — by a different group in a different country.

The load-bearing experiment is Jong, Azuma and Schaffer’s. Heart muscle cells from two- to three-day-old rats, forty-eight hours in medium with 5 mM β-alanine — which displaces taurine from its transporter and empties the cell.⁶

Two things out of that. It is not “the mitochondria decline”: ND5 and ND6 fall while COI is left alone, and COI is made by the same apparatus in the same mitochondrion. A general translation problem does not produce that picture.

And it really sits on translation rather than on reading the DNA: the ND5 protein falls while the ND5 mRNA does not change. The authors put that down to the tRNA^Leu(UUR), and it is the most direct indication in that paper that the step at issue sits on the transfer RNA.

One line in that table sits awkwardly against the human measurement. In the heart muscle cell complex III falls alongside complex I, to 35 % of control. In the Parkinson’s substantia nigra that does not happen — Schapira’s succinate cytochrome c reductase measures complexes II and III together, and it was unchanged. So the two signatures agree on complex I down and complex II intact, and differ on complex III.

The selectivity has been confirmed separately, and in a way that needs no protein staining. Shetewy and colleagues had the same cells respire on glutamate and malate, which runs through complex I, and on succinate, which bypasses it. On glutamate/malate they came out at 0.78 ± 0.08 of control (p < 0.05); on succinate at 0.98 ± 0.09. The mitochondria fragmented and caspases 3 and 9 were activated.³⁴

Then the human side, which is old and well repeated. Schapira measured NADH cytochrome c reductase at 4.36 ± 1.41 in nine controls and at 2.68 ± 1.01 in nine Parkinson’s patients, in nmol/min/mg protein — a fall of 39 per cent, p < 0.01 — and succinate cytochrome c reductase in those same two groups at 9.46 ± 3.01 and 9.59 ± 3.15 respectively. Citrate synthase and protein content were equal in both groups, so there were just as many mitochondria in that tissue.³²

One thing in there is not settled. By immunoblotting Schapira found no difference in the complex I proteins they could stain — Fe-S proteins of 30, 24 and 13 kDa, all three encoded in the cell nucleus and not in the mitochondrion; ND5 and ND6 were not among them. Flønes, by immunohistochemistry, did see less of a nucleus-encoded subunit. At subunit level the human picture does not add up.

And then what else collapses when taurine runs out. Jong, Ito and Schaffer took mice without a taurine transporter — a real deficiency inside the cell, not in the medium — and looked at the disposal service. Ubiquitinated protein piled up alongside a reduced 26S proteasome activity, and autophagy jammed: more Beclin-1 and more autophagosomes, but too few working autophagolysosomes. Treatment with mitoTEMPO, an antioxidant that works only inside the mitochondria, largely abolished the protein pile-up.³⁵

That last one settles the direction: the disposal failure is a consequence of oxidative stress out of the mitochondria and not a separate phenomenon. Anyone who knows the Parkinson’s literature will see the familiar trio standing here — faltering complex I, oxidative stress, and protein that no longer gets cleared — reached along one route: too little taurine inside the cell, in heart muscle at that.

Two things, both on the record, and the first is an objection from the people who have measured the complex I deficiency most precisely.

Flønes and colleagues counted complex I in individual neurons in eighteen people with idiopathic Parkinson’s and eleven matched controls, across several brain regions at once. The deficiency was everywhere in the Parkinson’s brain, including the cerebellum where nothing degenerates, it did not follow the anatomical distribution of the cell death, and outside the substantia nigra it did not track mitochondrial DNA damage. Their own conclusion is that it may therefore play no part in the degenerative process.³⁶

There is another reading of the same data. A brain-wide deficiency that does not follow the distribution of cell death is exactly what you would expect from a metabolic shortfall rather than a local injury: everyone carries the cut, and only the cell that cannot afford it goes under. The dopamine neuron of the substantia nigra happens to be the most expensive one compared so far — the highest basal oxygen consumption, the largest axonal arbor — and it demonstrably becomes less vulnerable once you make that arbor smaller.¹² It is then not the location of the shortfall that decides where cells die, but the location of the bill. That is an explanation and not a measurement.

That same 2018 paper holds something else that at first sight runs against the story: α-synuclein aggregation was in fact less common in complex I-poor neurons. Their follow-up splits the pathology into early and late and shows it is no contradiction. In eight people with idiopathic Parkinson’s, punctate, early inclusions sat in 10 of the 18 complex I-negative cells against 28 of the 176 positive ones (p = 6.3 × 10⁻⁵), while Lewy bodies sat in 0 of 18 against 23 of 176. Complex IV deficiency was associated with neither.³⁷ The authors read it as a double hit: cells with both are cleared first, and what you still see standing at the end with a Lewy body in it are the cells that survived.

The second counter-finding is about taurine itself. In the test tube, taurine speeds up the clumping of α-synuclein. Hegde and Rao tested five osmolytes on purified α-synuclein; glycerol, trimethylamine-N-oxide, betaine and taurine all four drove the protein into a partially folded intermediate, and it is that state which accelerates fibril formation.³⁸

The reach is limited — purified protein in a tube, and it is a property of osmolytes as a class rather than of taurine in particular. The concentration at which it flips is in their figure legend: 1.0 molar taurine. A cell holds millimolar amounts, and outside the cell there is 25 µM. It says nothing about a cell with a working disposal service. It has not been refuted, and it points the other way.

The absorption side first, because everything else hangs on it. Oral taurine is absorbed high up in the small intestine, with a plasma peak at an hour and a half; the transporter doing most of that work is the low-affinity, high-capacity PAT1 alongside the high-affinity TauT, and in rats absorption stayed not saturable across 10–997 mg/kg.^39,40,41 So you cannot “dose” taurine into your colon by taking more of it. The colonic route runs through bile: roughly 95% of conjugated bile acids are reabsorbed in the terminal ileum, and the remaining ±5% reaches the colon, where bacterial bile salt hydrolase cleaves the taurine off.⁴²

That this route really works this way has been shown most sharply through diet. Mice on a diet rich in milk fat produced more taurine-conjugated bile acid, upon which B. wadsworthia bloomed and Il10^−/− mice developed colitis. The decisive experiment: supplementing taurocholic acid did it, glycocholic acid did not.⁴³ So it is specifically the taurine conjugation that feeds the sulfur breathers.

And then there is the other side: taurine-driven sulfide is not simply bad. Taurine supplementation in mice strengthened the gut flora’s resistance to Klebsiella pneumoniae and Citrobacter rodentium, demonstrably through sulfide production.⁴⁴ The distinction is apparently one of quantity. In cultured human colonic cells the threshold sits at 50 µM: above it respiration falls, below it the cells oxidise the sulfide away themselves.⁴⁵ In colonocytes from human surgical specimens, sulfide inhibited butyrate oxidation but not glucose oxidation⁴⁶, and in mice sulfide breaks the disulfide bridges in mucin, opening up the mucus layer.⁴⁷

That bacteria breathing on taurine exist is solid: a taurine-respiring species in the mouse gut has been described that depends on other bacteria to release taurine from bile acids, and whose colonisation in turn increases that deconjugation.⁴⁸ In the human gut, B. wadsworthia is the best-known member of that guild — the same species that forms the fourth sulfur route in the fibre piece.

Since April 2026 there is a mechanistic explanation for the specificity Devkota could only observe. B. wadsworthia turns out to have its own secreted enzyme — a metal-dependent bile salt hydrolase outside the known enzyme family — with specific activity toward taurine-conjugated bile salts. It is conserved within the species and occurs in many other Desulfovibrionaceae. So the sulfur breather does not wait for another bacterium to free its substrate; it cuts it loose itself, and selectively.⁴⁹ This is a preprint without peer review, so no step in the account above rests on it — it merely explains neatly what had already been measured.

Against that stands the endogenous side. You make H₂S yourself, with CBS, CSE and 3-MST, and it behaves as a gasotransmitter that regulates your antioxidant defence, your vascular function, your inflammatory response and your mitochondria — and it happens that dysregulation of that transsulfuration pathway has been described precisely in Parkinson’s, Huntington’s and Alzheimer’s.^50,51

And then the finding that shows most sharply how much this is a story about place: H₂S donors are neuroprotective in the same Parkinson models. NaHS restored motor function in 6-OHDA and rotenone models, preserved TH-positive neurons in the substantia nigra and inhibited microglial activation via NF-κB; the slow-releasing donor GYY4137 did the same in the MPTP model.^52,53 That is exactly the mechanism by which taurine worked in PART II. The same molecule, the opposite outcome, and the difference lies in place, dose and regulation.

In humans that step has been measured once, and not in Parkinson’s research. In the prehypertension trial, plasma taurine in the taurine group rose from 108.3 to 142.3 µmol/L and plasma H₂S from 43.8 to 87.0 µmol/L, and the change in blood pressure correlated negatively with both; in the placebo group neither happened.⁸ Two changes moving together are not a demonstrated route, and a blood pressure trial is no model of this disease. What is there is a measurement in humans in which swallowed taurine raised the sulfide while the outcome went the right way.

Biochemically the two are literally related: cysteine catabolism splits into an oxidative route via cysteine dioxygenase to taurine and sulfate, and a non-oxidative route via CBS, CSE and 3-MST to H₂S.^51,54

In the rotenone model they measured that shift directly as well. Taurine largely undid the dysbiosis the toxin had caused, and at the same time gut motility came back, and the barrier function of both the intestine and the blood-brain barrier, and LPS, IL-1β, IL-6 and TNFα fell in plasma.²² That is the gut-brain axis in a single experiment, but without a faecal transplant or germ-free control it again stays unresolved whether the gut carries the protection or merely moves along with it.

There is a second thread between fibre and this route, and it is counter-intuitive. In rats, pectin dose-dependently raised bile acid excretion, but that increase consisted entirely of glycine-conjugated bile acid while the taurine conjugates fell — precisely the conjugation that, per Devkota, does not feed Bilophila. The explanation is that the raised bile acid excretion draws on the hepatic taurine store, after which the liver switches to glycine.⁵⁵

This is visible in humans too, though cross-sectionally rather than by intervention. In 36 vegans against 36 omnivores — more fibre, less fat — serum primary and glycine-conjugated bile acids were in fact higher and all faecal bile acids lower. In the pattern analysis fat intake correlated positively and fibre intake inversely with the bile acids; that pattern explained 47.4% of the variance.⁵⁶

But the comparison with Ide only half holds, and that emerges only from the full text: serum taurine-conjugated bile acids did not differ between vegans and omnivores. What was measured is an increase on the glycine side, not a shift away from taurine. In the rat the taurine conjugates did fall; in these people they did not. Fibre apparently does not remove the sulfur breathers’ food. Further: cross-sectional, not randomised, and vegans differ in more than fibre alone.

The individual foods from that pattern analysis say less than they appear to. Loading positively on the bile acid pattern were not only processed meat, fried potatoes and margarine but also fish and coffee — so it is not a healthy-versus-unhealthy axis. And what loaded most negatively was muesli, which in a German cohort of 72 people probably flags the whole group rather than doing anything itself — though that is a reading, not a measurement. Loadings in a pattern analysis are not dietary advice. What the study did find are associations: fat intake positively, fibre intake inversely.

Fibre moreover raises the flora’s deconjugation capacity, and that works the other way: whole food intake went together with more bile salt hydrolase activity and with fewer conjugated bile acids in stool, supported by a fibre-free intervention in healthy people.⁵⁷ More deconjugation means the taurine is in fact freed faster. So fibre does not remove the substrate — it changes who is standing ready to eat it. That is the same conclusion as above, now from a third direction.

The catch belongs right alongside it, because it is in the same study: when those rats were given taurine as well, glycine conjugation was all but abolished and taurine conjugates dominated again. Fibre plus taurine simply yields taurine-conjugated bile again. The protection fibre offers therefore lies not in less substrate but in fewer sulfur breathers to eat it. (Rat, 1989, and one kind of fibre — pectin, not acacia; this is mechanism, not a dosing recommendation.)

That last part has been measured in humans once. Patients with functional gut disorders were randomised, double-blind, to either a prebiotic — 2.8 grams a day, containing 1.37 grams of galacto-oligosaccharide — or a low-FODMAP diet. On the prebiotic, B. wadsworthia fell. On the low-FODMAP diet it rose.⁵⁸ The p value is .050, right on the line. It is one trial, with a galacto-oligosaccharide and not with acacia fibre. But both arms point the same way as the mechanism above: where there was more to ferment the sulfur breather grew scarcer, and where there was less to ferment it grew more numerous.

For completeness, the Parkinson’s part of this axis. In mice with an MPTP model, Lactobacillus, Lachnospiraceae and Adlercreutzia were reduced and Aerococcus, Staphylococcus and Ruminococcaceae raised, while serum taurine had clearly fallen; adding taurine (150 mg/kg daily, four weeks) restored movement, protected dopamine neurons and reduced microglial activation.²⁵ The weak link right alongside it: no faecal transplant was done. Both fall together, and which one is driving is unresolved.

This is the DeNoPa cohort: 30 drug-naïve, newly diagnosed patients and 30 matched controls, followed with measurements at baseline, 24 and 48 months. Three taurine-conjugated bile acids were positively associated with the UPDRS-III score. In a separate population cohort, sulfated taurolithocholic acid was instead protectively associated with the incidence of Parkinson’s — so even within this family the metabolites do not point the same way.⁶²

In the same work Bilophila wadsworthia and Akkermansia muciniphila were raised in Parkinson’s, with consequences for sulfite and H₂S production, and cystathionine accumulated in the sulfur route.⁶² That is the same axis as in PART IV — and it makes the most natural reading not “taurine is bad” but these guts convert taurine into sulfide instead of passing it on.

This is an interpretation consistent with the known biochemistry, not a demonstrated link. The cohort was not designed to establish direction, and nobody measured whether sulfide was actually higher in these people. Anyone who prefers to read the finding as crossing out the whole taurine story is, on these data, equally entitled to do so.

There is moreover a confound that weighs heavier than the gut reading, and it comes from outside Parkinson’s research entirely. The glycine:taurine ratio of conjugated bile acids shifts towards taurine in illness generally. In 70 ARDS patients that ratio was substantially lower than in healthy controls — 1.79 against 3.80 — and specifically because of an increase in the taurine conjugates; it correlated with the SAPS II severity score and with survival, and the authors explicitly read it as an adaptive response to illness.⁵⁹ In chronic liver disease, glycine:taurine ratios are so stage-specific that diagnostic models built on them reach an AUC of 0.91 to 0.97 in a cohort of 1883 participants.⁶⁰

Set that beside the finding: in DeNoPa the taurine conjugates tracked the UPDRS-III — a severity measure. A marker that moves with severity in illness generally also moves with severity here. That is exactly what you expect of a consequence, and it cannot be told apart from a cause in a design without an intervention.

And then the question that actually matters to a reader: how solid is the bridge between taking taurine and taurine-conjugated bile acid? Weaker than you might think, but not nil. In humans, hepatic taurine concentration is the main determinant of the percentage of bile acids conjugated with taurine, and as little as 250 mg of intraduodenal taurine raised that percentage by 2.5 to 10 percentage points in 7 of 10 subjects within two and a half hours. That same study explicitly adds that the deviation is small and that the composition of the circulating bile acid pool does not change acutely.⁶¹

Taken together: this is a cross-section without an intervention, in which the exposure (taurine-conjugated bile acid) is at best indirectly and weakly coupled to what someone swallows. As evidence against supplementation it is therefore weak; as an indication that a sulfur-heavy gut shows up in the bile acid profile it weighs more.

The limitation first, because it is real. Transport across the blood-brain barrier runs via the taurine transporter TauT (SLC6A6) and is saturable. At raised plasma levels the transporter is downregulated, and under oxidative stress and in disease states the transport works less well — exactly the circumstances in which you would want it.⁶³ That bounds how much gets in. It does not close the route.

That the route is open has now been measured in humans. In blood from the cerebral venous sinus against arterial blood, taurine ranked with glucose and hypoxanthine among the compounds with high net uptake by the brain — out of 1,365 metabolites and 140 lipids.⁶⁴ The participants had sinus stenosis or thrombosis, because that vein is only reachable by catheter; it was neither a supplementation nor a Parkinson’s study.

And oral intake demonstrably reaches the central nervous system. In an open-label trial in SSADH deficiency, spinal fluid taurine was elevated during oral treatment, in the four participants from whom fluid was available.⁶⁵ That trial was clinically negative — no cognitive gain, and the TMS outcome ran against the hypothesis. Different disease, different endpoint; what follows from it here is solely the transport fact.

That spinal fluid and brain tissue are not the same thing shows up in the animal. Rats given taurine for fourteen days held the same total amount of taurine in their brains.⁶⁶ Probably the throughput rises while the pool inside the cell stays regulated — which is what you would expect of a saturable transporter. Anyone hoping to fill their brain with taurine is hoping for something no one has measured.

At high doses there is even a human chain from capsule to molecular target — the MELAS trial from PART III, in which spinal fluid taurine went from 11 to 42 micromoles per litre and the modification on the mitochondrial tRNA rose in five of the ten people. The four reasons not to read too much into it are there as well.

What remains is therefore a narrower uncertainty than “does it get there”: how much brain exposure roughly 850 mg a day produces in someone with Parkinson’s, and whether that is enough. No measurement of that exists. Beyond it, part of this story plays out outside the brain anyway — the gut route from PART IV and systemic anti-inflammatory action — but that is reasoning, not measurement.

That claim circulates widely, and it is too simple. When they infused taurine straight into the striatum of rats, extracellular dopamine rose — 2.5-fold at 150 mM over two hours, and 10-fold at 450 mM, but then only in the first half hour. When they put the same taurine a little further along, in the substantia nigra, striatal dopamine fell instead.⁶⁷

The effect therefore depends on where the compound ends up, at concentrations that have nothing to do with swallowing. “Taurine raises dopamine” is not a property of taurine. Nowhere does the story here run through dopamine; it runs through inflammation.

Unlike Parkinson’s, here there is straightforward randomised evidence in humans. A meta-analysis of 25 trials with 1024 participants found a fall in systolic pressure of about 4.0 mmHg and in diastolic of 1.5 mmHg, plus a dose-dependent improvement in fasting glucose.⁹ To be precise: −4.00 mmHg (95% CI −7.29 to −0.71), −1.51 mmHg (−2.48 to −0.54) and −5.88 mg/dl fasting glucose (−10.75 to −1.02), at doses from 0.5 to 6 g/day and durations from five days to a year. There is a second meta-analysis you often see cited alongside it, but that is not an independent confirmation: both from Tzang and colleagues, both from 2024, and a systolic estimate identical to three decimal places. That one counts 20 trials and 808 participants and comes out at −1.44 mmHg diastolic.⁶⁸ In heart failure, clinical doses of 500 mg to 6 g per day have been used, though the systematic review that collected them found no significant association for ejection fraction or stroke volume.⁶⁹

The broadest and most recent meta-analysis — 34 trials — arrives at comparable values: −4.38 mmHg systolic, −2.54 diastolic, and −5.90 mg/dl fasting glucose.⁷⁰ Only the abstract of it was available.

The trial this part opens with comes with a little more detail. Of the 120 randomised participants, 86 were analysed, 44 on placebo and 42 on taurine. The fall of 7.2 against 2.6 mmHg is the clinic measurement, and that is the measurement on which the comparison between the two groups reached significance, at 8 and at 12 weeks. Measured across a whole day with an ambulatory monitor, the upper reading fell 3.8 mmHg against its own baseline while the placebo group went up 0.3, and the lower reading 3.5 against 0.6. Smaller, then, and on a different kind of test. Those who started highest fell furthest: with a high-normal starting pressure (130–139/85–89) it came down 10.1 mmHg against 3.0 in the low-normal group.⁸

A formal upper limit does not exist. No systematic pattern of adverse effects surfaced in humans, so there was nothing to base a NOAEL on, and Shao and Hathcock arrived instead at what they call an observed safe level — 3 g/day for healthy adults, which is to say the highest dose at which they saw nothing happen.¹¹ The Norwegian regulator went into more detail and assessed individual supplement doses: 750, 800, 900 and 1000 mg a day it considers unlikely to harm adults, while 2000 mg a day may represent a risk.³

Two things on that. The 850 mg from the animal-dose conversion falls exactly inside the range that was assessed directly — more useful than a distant ceiling. But the margin is not unlimited: between that dose and the point where VKM sees a risk lies a factor of two. And the Japanese medicine at 3 grams a day sits above what VKM holds for supplements. That is not a contradiction — a prescription drug is supervised and a tub from the chemist is not — but the two situations are not interchangeable either.

And there is a real deficiency side. Plants contain no taurine; vegans had a plasma level of 45 against 58 µmol/l in omnivores, and excreted far less of it in urine: 266 against 903 µmol per day.⁴

The authors themselves take it that the body probably becomes more frugal and makes enough. So it is a lower level rather than a demonstrated harmful deficiency: people eating plant-based sit consistently lower, and whether that harms them has not been measured.

A Canadian group measured circulating taurine in 137 men aged 20 to 93 and found no association at all: not with age, and not with muscle mass, strength, physical performance or mitochondrial function.⁷¹

Levodopa is structurally related to phenylalanine and tyrosine and is transported by LAT1 (SLC7A5–SLC3A2), both across the gut wall and across the blood-brain barrier. The seven large neutral amino acids — leucine, isoleucine, valine, phenylalanine, tyrosine, tryptophan and methionine — compete for the same seats, which is why protein intake is timed around dosing.⁷⁴

Taurine is not a large neutral amino acid and not a substrate of LAT1; it uses TauT (SLC6A6).⁶³ On that basis, competition at the LAT1 gate is not to be expected. That is an inference from transport pharmacology, not an interaction study. I found no research measuring taurine and levodopa absorption together.

The reverse has been measured, and there is an open question here: levodopa use goes together with lower plasma taurine, in proportion to the cumulative dose.¹⁸