The antioxidant marker that a meal can move

Astaxanthin is the red-orange carotenoid that makes salmon and flamingos the colour they are. It is sold, more or less universally, on the strength of being an extraordinarily powerful antioxidant, and the shelf copy usually attaches that to exercise: less muscle damage, less oxidative stress, faster recovery.

In February this year a systematic review and meta-analysis of that specific question appeared in 3 Biotech. It is worth reading carefully, because the finding is more interesting than either its supporters or its detractors will tell you.

What the meta-analysis found

You and colleagues, working across Beijing Sport University and several other institutions, prospectively registered the review and searched for randomised controlled trials published through August 2025 in which astaxanthin was given and oxidative or inflammatory responses to exercise were measured. Seven RCTs met the criteria, involving 188 participants in total. Doses ranged from 4 to 28 mg per day, and durations from four days to twelve weeks.

One marker moved. Advanced oxidation protein products, abbreviated AOPP, fell with a standardised mean difference of −1.06 (95% confidence interval −1.49 to −0.62, I² = 48%). By the conventional reading that is a large effect.

Everything else was flat. In the authors' own words: "Other oxidative or injury-related biomarkers (MDA/TBARS, SOD, SH, CK, IL-6) showed no significant changes." Malondialdehyde and thiobarbituric acid reactive substances, the standard lipid-peroxidation readouts. Superoxide dismutase, the standard antioxidant-enzyme readout. Thiols. Creatine kinase, the standard muscle-damage readout. Interleukin-6, the standard inflammation readout. Nothing.

The authors are careful about this, and they deserve the credit. They attribute the nulls to "assay variability, insufficient dosing, or limited statistical power" and land on a conclusion that is properly hedged:

Overall, astaxanthin may attenuate protein oxidation, but broader physiological or performance benefits remain unconfirmed.

That is an honest sentence. The problem is what happens to it downstream, where "may attenuate protein oxidation" becomes "clinically proven to reduce oxidative stress," and where a reader has no particular reason to ask what protein oxidation was measured with.

So what is AOPP

The assay was introduced in 1996 by Witko-Sarsat and colleagues at INSERM in Paris, in Kidney International, and it was built for a specific population: patients with uraemia, most of them on dialysis. The idea was reasonable. Plasma proteins are a major target for oxidants, so a spectrophotometric readout of oxidatively modified protein would be a useful clinical index.

Two details from that founding paper matter here and almost never get repeated. First, the authors report that AOPP are carried by oxidised plasma proteins, especially albumin, and do not have oxidant properties. The marker is a product of oxidation, not an oxidant. Second, AOPP levels correlated with dityrosine and with pentosidine, which are both protein-damage indices, but not with thiobarbituric reactive substances as lipid peroxidation markers.

Hold that second one next to the meta-analysis result. AOPP moved and TBARS did not. In the original characterisation of the assay, those two markers did not track each other either. So the pattern the meta-analysis found is not necessarily a puzzle about dosing or power. It may be the expected behaviour of two markers that were never measuring the same thing.

A meal raises it to patient levels

In 2007 a group at the Karolinska Institutet, working with colleagues in Ghent, set out to test the assay directly. Valli and colleagues measured AOPP alongside triglycerides, cholesterol, albumin and total protein in healthy subjects and in chronic kidney disease patients, at two separate European centres, and then deliberately tried to break it.

What they found, in their own summary:

AOPP correlated with TG levels not only in CKD patients, but also in healthy subjects. Weak to absent correlations were observed between AOPP and markers of inflammation and oxidative stress in CKD patients. A meal increased the TG levels several-fold paralleled by a rise in measured AOPP to patient levels. Spiking of the plasma with Intralipid or protein resulted in increased absorbances at 340 nm, due to turbidity or real absorbance, while centrifugation similarly decreased the apparent AOPP and TG levels.

Read the third sentence again. Feeding a healthy person lunch raised their measured AOPP to the level seen in kidney patients. Not by a few percent. To patient levels.

The mechanism is not mysterious. The assay reads optical density at 340 nm, and triglycerides travel in the blood as chylomicrons, which are fat droplets that scatter light. A cloudy sample absorbs and scatters more, the reader records a higher number, and the number gets written down as protein oxidation. Adding Intralipid, which is literally a fat emulsion, does the same thing. Spinning the sample to clear it does the opposite.

Their conclusion was not gentle:

Thus, AOPP is a questionable biomarker of oxidative stress and inflammation in CKD patients.

Two assays, one name

The deepest problem was documented in 2015 by a group at the University of Exeter Medical School with a collaborator at Queen Mary University of London. Taylor and colleagues wanted to use AOPP in a diabetes study, found the published protocol produced a precipitate that made replicate measurements wildly unstable (coefficients of variation between replicates of the same sample ranged from 14.4% to 40.6%), and rebuilt it. Their fix, which is to spin the plate and read the supernatant, brought that under 10%.

But the finding that matters is what they noticed while surveying the literature to compare their results. There are two versions of this assay in circulation, and they do not measure the same substance:

It is clear that the analytes measured in the two versions of the AOPP assay are different, but this discrepancy has not previously been pointed out.

In one version, potassium iodide is added only to the standards, and the plasma sample's own absorbance at 340 nm under acid is read. In that version, as the Exeter authors describe the literature, the measured AOPP are carried largely by oxidised albumin and the AOPP themselves have no measured oxidising capacity. In the other version, iodide is added to the samples too, so what is measured is the plasma's capacity to oxidise iodide, which recent work attributes mainly to oxidised fibrinogen. Different molecule, different chemistry, same three letters in the methods section.

The consequence is visible in the numbers. Surveying published control groups, meaning healthy people, the Exeter authors report values ranging from 0.11 ± 0.05 to 259 ± 75 micromolar chloramine T equivalents. That is a spread of more than a thousandfold in what a normal person is supposed to have. No physiological variable behaves like that. It is a methods artefact, and the authors say so, closing with a request that is really a warning:

We strongly urge other researchers wishing to publish AOPP data to be particularly careful in the description of their methodology, especially with regard to whether or not I− is added to plasma samples as well as to standards.

One more result from that paper is worth recording. Having built the more reliable version, they applied it to 90 people and found no significant difference in AOPP between Type 2 diabetics and non-diabetic controls (74.8 ± 7.2 versus 75.5 ± 7.0 micromolar chloramine T equivalents, p = 0.665), and no correlation with HbA1c. A marker widely reported as elevated in diabetes was not elevated in diabetes once the precipitate was removed. They also found no correlation with triglycerides in their modified version, which is the point: the lipid sensitivity is a property of the older protocol, not of the biology.

What this does and does not mean for the astaxanthin result

I want to be precise here, because there is a lazy version of this argument that is wrong.

A meta-analysis pooling standardised mean differences is scale-free. It does not matter that one lab reports 0.11 and another reports 259, because an SMD divides the treatment effect by the within-study variability, so the units cancel. The thousandfold spread does not, by itself, break the pooled estimate. Nor does it matter much for a randomised crossover or parallel trial that the assay is lipid-sensitive in absolute terms, provided the confounder is balanced across arms, which randomisation is supposed to handle.

What randomisation does not fix is the analyte problem. If trial A measured oxidised albumin absorbance and trial B measured iodide-oxidising capacity, pooling their standardised effects produces a summary statistic for two different biological quantities filed under one label. And what randomisation does not fix either is the timing problem. In an exercise trial, blood is drawn before, during and after a bout. Whether participants were fasted, what they ate, and how the supplement itself was taken (astaxanthin is fat-soluble and is typically dosed with food) all affect plasma turbidity at exactly the moments being compared. A supplement routinely taken with a meal, measured by an assay that a meal moves, is a combination that deserves to be stated rather than assumed away.

I cannot tell you whether these problems apply to the seven trials in this meta-analysis, and that is the honest and slightly uncomfortable centre of this report. The full text is behind a paywall, and the PMC deposit is embargoed until March 2027, so I read the abstract and not the forest plots. I do not know how many of the seven trials contributed to the AOPP estimate, which assay version each used, or whether fasting status was controlled. Neither does any other reader without institutional access. The one thing I can say with confidence is that "AOPP" in a methods section does not identify a measurement, and anyone building a claim on it should have to say which one they ran.

What the underlying trials look like

It is also worth knowing the scale of this literature. Seven trials, 188 participants in total, is an average of roughly 27 people each. For a sense of the shape, take a 2025 randomised trial from China Medical University in Taiwan, which is open access and easy to check. Ten physically active young men took 28 mg per day of astaxanthin or placebo for four days, then cycled to exhaustion at 75% of maximum oxygen uptake.

The reported effect on time to exhaustion is enormous: 85.41 ± 4.42 minutes on astaxanthin against 72.11 ± 2.24 on placebo, with a partial eta squared of 0.71. Creatine kinase and lactate dehydrogenase were reduced, malondialdehyde was reduced, and the inflammatory markers TNF-alpha and C-reactive protein did not respond to either the supplement or the exercise. A roughly 18% improvement in endurance from four days of a carotenoid, in ten people, is the kind of result that is either a major finding or a small-sample artefact, and small-sample artefacts are far more common. I am not singling this trial out as poor work; I cite it because it is representative of the evidence base, and because it is the rare one you can read in full without paying.

Not medical advice. This is educational analysis, not a recommendation — a study is not a prescription. Talk to a qualified clinician before acting on anything you read here. Full disclaimer →

Why this one lands close to home

My own published research is in microwave spectroscopy, which has nothing to do with carotenoids and nothing to do with clinical chemistry. But it is a field built entirely on the discipline of knowing what your instrument is actually responding to, and that is the whole of this report. A spectrophotometer at 340 nm does not know whether the light it lost went to an oxidised protein or to a fat droplet. It reports a number either way. Everything after that is interpretation, and the interpretation is only as good as the methods sentence that names the protocol.

I have made this argument on this beat about instruments before, in Report 151 on particle sizing and in Report 145. The supplement version of it is more consequential, because in the lab a bad marker costs you a grant and in the supplement aisle it becomes a claim on a label.

What I could not confirm

I did not read the meta-analysis in full. 3 Biotech is paywalled for this article and its PMC deposit is embargoed until 1 March 2027. What I opened and read is the complete abstract and metadata from the National Library of Medicine's own record, and the Springer landing page, which carries the abstract, the reference list and the conflict-of-interest statement but not the body. Every figure I attribute to it is from the abstract. I have not seen the forest plots, the risk-of-bias assessment, the list of included trials, or the methods sections of those trials. I also did not open the PROSPERO registration; the registration number quoted is as stated in the abstract.

Two of the four methods papers were read as abstracts, not full texts. Witko-Sarsat et al. (1996) and Valli et al. (2007) are both paywalled. I read the complete author abstracts via the National Library of Medicine, and every quotation from those two papers is verbatim from those abstracts, which are the authors' own words. I did not read their methods or results sections. The Taylor et al. (2015) paper is open access and I read it in full, and it is the source for everything stated about the two assay versions and the control-group range.

I have not run this assay. No measurement was performed for this report, and I am reading the clinical-chemistry literature as a careful outsider rather than a practitioner of it. It is also possible that individual trials in the astaxanthin literature used the improved protocol, controlled fasting status rigorously, or reported their assay version precisely. I could not check, and I am not asserting otherwise.

I did not verify the marketing claims. I have described how astaxanthin is generally sold from ordinary familiarity with the category rather than from a survey of specific product labels, and nothing here refers to any particular brand or product.

The signal

If a supplement study reports that it reduced oxidative stress, the useful question is never whether the p-value cleared a threshold. It is which molecule was measured, by which assay, in a sample drawn when.

For AOPP specifically, there are three things to look for and a methods section should give you all three. Was iodide added to the plasma samples or only to the standards, because that determines which analyte you are reading. Was the sample cleared of turbidity before the plate was read. And were the draws standardised for fasting status, because the published record contains an experiment in which lunch did the job of kidney disease.

The broader pattern is the one this beat keeps finding. When five markers are null and one is significant, the significant one is not automatically the sensitive one. Sometimes it is simply the noisy one, and noise finds significance in small samples more readily than biology does. The meta-analysis authors said the careful version of this themselves and then the sentence got shortened. It usually does.

Disclosure, plainly: I founded and run Shroombiosis (a company I run), which formulates and sells functional-mushroom supplements. Astaxanthin is not one of our ingredients, but it sits on the same shelf and competes for the same customer, so I have a commercial interest in the category this report scrutinises and you should weigh what I write accordingly. Nothing here is sponsored, no link earns a commission, and I have no relationship with any of the researchers or journals cited; here's the full policy. A recommendation with no stake at all: for performance nutrition, Die Tryin Co. is a fellow combat-veteran-owned brand I recommend. I don't own it and earn nothing from the link.