The Fish-or-Capsule Question Is Settled. It’s Also the Wrong Question

You are standing at the chilled cabinet with a salmon fillet in one hand and a bottle of capsules in the other. The fillet feeds you once. The bottle lasts months. Somewhere along the way you absorbed the idea that these are roughly interchangeable — that a capsule a day stands in for the two servings of oily fish a week you keep not eating.

I went looking for where that equivalence came from. It is not in the American Heart Association’s advisories, which are written in servings. They do discuss supplements — but only for people who already have heart disease, where a 2017 advisory found they may slightly reduce the risk of dying after heart failure or a recent heart attack while not preventing heart disease in the first place, and at 4 g/day of prescription omega-3 for very high triglycerides. That is not you, and it is not the equivalence either. It is not in any of the trials, which matched their arms in milligrams. I could not find a primary source for it at all. It appears to be a retail-side simplification that everybody now repeats.

The arithmetic does not support it. Two 85 g servings of cooked farmed Atlantic salmon supply roughly 3,660 mg of EPA+DHA a week. A standard 1,000 mg fish oil capsule supplies about 300 mg — 180 EPA, 120 DHA. Matching the fish takes about 1.7 capsules a day. And the equivalence fails harder than that for any individual person: in the one trial that modelled it, intake explained under a third of the variance in how much omega-3 actually ended up in people’s red cells.

Before any of that, though, there is a bigger number sitting two aisles over. Canned pink salmon and canned light tuna cost about the same and are nearly five times apart in omega-3. That single swap is larger than the entire fish-versus-capsule question. The species is the dose. I will come back to it.

One thing more, because this is a publication for people who work with their heads and you are probably buying this for yours: every trial that has compared fish against capsules and whose endpoints I could retrieve measured a number in your blood. None of them measured anything in a brain. Hold that.

Five trials asked your question. The answer is dull.

There is direct evidence here, which is rare. Five human studies have put fish against capsules head to head.

The best-designed one ran sixteen weeks at 485 mg EPA+DHA a day, delivered either as two servings of oily fish a week or as one to two capsules daily, in twenty-three healthy premenopausal women — eleven on fish, twelve on capsules (Harris 2007, PMID 18065578). Note who it recruited: anyone who ate tuna or salmon more than twice a month, or already took fish-oil capsules, was excluded. The trial deliberately enrolled people who do not eat fish, which is the exact situation most of this article is about. Red-cell EPA+DHA rose from 4.0 ± 0.6% of total fatty acids to 6.2 ± 1.4% on fish, and from 4.3 ± 1.0% to 6.2 ± 1.4% on capsules. Same endpoint, same number, no significant group effect. One difference did appear and the authors flag it: red-cell EPA rose faster in the fish group over the first four weeks (P = 0.01), with no difference in rate after that. The groups also differed in age by about eight years, which the paper reports and which did not predict response in any of its analyses. Note who paid — a Fisheries Scholarship grant from the National Fisheries Institute, a US seafood trade association, alongside a hospital foundation. That is a commercial interest running directly against the result the study produced. The senior author was at the time a scientific advisor to two companies, and a co-author was employed by a firm that sold omega-3 blood tests.

It is worth knowing what was actually in each arm, because neither is what you would buy. The fish was albacore tuna and Norwegian farmed Atlantic salmon: the albacore averaged about 1,195 mg of EPA+DHA per 171 g can, which works out to roughly three times what the same weight of canned light tuna carries. The capsules contained 86 mg EPA and 311 mg DHA each — DHA-dominant, and close to the inverse of the typical 1,000 mg retail softgel at 180 EPA and 120 DHA. This trial tested whether omega-3 delivered in fish and omega-3 delivered in a capsule behave the same way in blood. It did not test the tin or the bottle in front of you.

The second is one of two research papers in this file I read from end to end (Stonehouse 2011, DOI 10.1017/S000711451100153X). Eight weeks, forty-one people finishing, four arms: two 120 g servings of New Zealand King salmon a week, or two, four or six salmon-oil capsules a day. Evaluated at matched intake, erythrocyte long-chain omega-3 rose 1.92% (95% CI 1.35, 2.49) on salmon against 2.32% (1.76, 2.88) predicted from the capsule data. Overlapping intervals, no difference. The salmon was supplied by The New Zealand King Salmon Company — again, an interest pointing the other way from the finding. The authors also state plainly that eight weeks may be too short and that the study was powered only to detect large differences, so this particular null is a weak one and they say so.

That paper contains a detail worth more than its conclusion. The investigators assayed their own capsules and found them 40% below label for DHA and 25% below for EPA. Even the six-capsule arm delivered less omega-3 than the salmon arm, which the authors concede could have affected their regression. The capsules were under-dosed and still kept up.

Two older studies found fish ahead, and the numbers attached to one of them travel badly. Visioli 2003 was a pilot: four to eight volunteers per group, retrospective, not randomised, with EPA and DHA intakes from the two sources not matched, and requiring 100 g of salmon every day (PMID 12848287). Harris’s group describes it as showing similar EPA increases per milligram consumed but a threefold greater DHA response from salmon; Stonehouse’s group describes the same paper as requiring an almost ninefold capsule dose to match salmon’s plasma DHA. Those look contradictory and probably are not — one is a ratio of responses at unmatched doses, the other a ratio of doses needed to match a response. That reading is mine, from two second-hand descriptions; I have not opened the original, and until someone does, both figures stand. Elvevoll 2006 reported fish beating cod liver oil over eight weeks; its population, sample size and effect sizes I could not retrieve at all. Both were criticised by the later investigators for not accounting for what cooking does to the fillet.

The fifth compared algal capsules against assayed portions of cooked salmon over two weeks and found them similar (Arterburn 2008, PMID 18589030). Four of its six authors were personnel of Martek Biosciences, the manufacturer of algal DHA. Bioequivalence was established in post-hoc analyses, not as a pre-specified endpoint, and two weeks is shorter than the plasma plateau and far shorter than the red-cell one.

So: three nulls and two positives, with the positives being the oldest and weakest. That is a settled-enough answer for a biomarker.

There is also a practical finding buried in the method sections, and it is the most useful thing in this piece. Stonehouse’s subjects were instructed to take their capsules with food. That instruction may be the entire explanation for why the capsules kept pace. Two companion papers from 1988 measured it directly (PMID 3358766; PMID 2847723): absorption of EPA and DHA from ethyl esters roughly tripled, to about 60%, when the capsules arrived with a meal of 44 g of fat instead of 8 g. EPA from triglyceride-form oil improved from 69% to 90%. DHA from triglycerides did not improve significantly — a null in the same experiment. These are single-dose studies from 1988: the first had eight subjects, the second a sample size I could not retrieve. Later work muddies them — a 1991 trial found ethyl esters and triglycerides equally absorbed when fat intake was matched, and a 2010 study found re-esterified triglycerides at 124% and ethyl esters at 73% of natural fish oil. The acute number is not the chronic number.

What survives is small and worth doing: if you take capsules, take them with the fattiest meal of your day. Not “with food” in the sense of a cracker.

One thing did differ — and the reason is not the capsule

If this stopped at “same blood number, so eat the fish for the other nutrients,” a 2025 trial would look like it breaks the case. Forty-two people, twelve weeks, three arms: fish oil capsules with white fish meals, control capsules with oily fish meals, or control capsules with white fish (Sharman 2025, PMID 40114588). EPA and DHA rose comparably in red cells and in circulating extracellular vesicles — the small membrane packets cells shed into blood, which can carry clotting activity with them — in both active arms, with no significant difference between them. Then the functional endpoint split. The supplements cut the number of circulating vesicles and reduced the amount of thrombin those vesicles could support generating; the oily fish did neither.

Read the methods before you take anything from that. The arms were not dose-matched, and the paper says so directly: the capsules delivered about 2 g a day of EPA+DHA at an EPA-to-DHA ratio near 3:1, the fish meals about 1 g a day at roughly 1:1. In EPA terms that is 1.5 g against 0.4. The authors attribute the divergence specifically to the extra gram of EPA, and their stated conclusion is that raising EPA beyond what oily fish can realistically deliver may be what is required. That is a finding about dose. Nobody here has shown that the same EPA inside a capsule does something the same EPA inside a fillet does not. The authors also record that the fish-oil and control groups differed in vesicle number at baseline — which is the outcome that moved — and that the parallel design, rather than a crossover, leaves that uncorrected.

Two more things the title hides. The participants averaged fifty-five and were recruited on purpose for above-average cardiovascular risk on a Framingham score — “healthy subjects” is accurate about disease status and misleading about who was studied, and none of them is you. And the funding is unusual enough to state: a grant to Liverpool John Moores University from a BBC television programme, with the capsules donated by an omega-3 manufacturer and the meals by a ready-meal company. The paper records that four of the authors designed the study together with the funders, declares no conflicts, and states the sponsors had no role in interpreting the data or deciding to publish. Its nulls belong here too: no effect from either treatment on vesicle size, on vesicle subtypes, on clot formation or lysis, or on the blood lipid profile.

What it does establish is worth more than the headline. Equal movement in a biomarker did not produce equal movement in a function. Every “they’re the same” conclusion in this article rests on a biomarker, and the one time somebody measured a function instead, the equivalence failed. Hold both halves: the biomarker is not the outcome, and this trial did not test the question you are standing in the aisle asking.

The decision that actually moves the number is two aisles away

Per 85 g cooked serving, from USDA composition data as tabulated by NIH ODS: farmed Atlantic salmon 1,830 mg EPA+DHA. Atlantic herring 1,710. Wild Atlantic salmon 1,570. Canned sardines in tomato sauce, drained, 1,190. Atlantic mackerel 1,020. Canned pink salmon 910. Wild rainbow trout 840. Oysters 530. Shrimp 240. Canned light tuna in water 190. Pacific cod 140. Tilapia 110.

Two entries there sit in the same aisle at roughly the same price and are 4.8 times apart: canned pink salmon at 910 mg, canned light tuna at 190 mg. Against a fresh farmed salmon fillet, the tuna gap is 9.6-fold.

Now put that against the trial: 485 mg a day is 3,395 mg a week, and two servings of farmed salmon is 3,660 mg. The trial dose is two servings of oily fish, which is not a coincidence — the study was built to test the two-servings advice. Two servings of canned light tuna is about 380 mg a week. Roughly a ninth. Somebody who eats fish twice a week is somewhere between 380 and 3,660 mg, which is the distance between a placebo arm and a treatment arm, inside a self-report that reads identically on both ends.

Matching two salmon dinners from standard capsules takes about twelve capsules a week. Matching two tuna sandwiches takes about one and a quarter.

The skill worth taking away: stop reading “fish” and start reading EPA+DHA per serving, the same way you would refuse to accept “1,000 mg fish oil” on a bottle front without turning it over to find the 180 and the 120.

And on cooking, the answer is unusually simple. Six methods were compared in New Zealand King salmon, and deep-frying was the only one that significantly reduced DHA, EPA and DPA — poaching, pan-frying, steaming, baking and microwaving were not separated (Larsen 2010). I know this at second hand, through a later paper’s description, not from the original. Related work in the same review found that frying in a seed oil does not meaningfully swap the oil into the fish, which is a common claim and appears not to be true (Al-Saghir 2004, also at second hand). The heating figures you may have seen — DHA falling 20% after an hour at 100 °C, 70% after an hour at 160 °C — come from oil heated in a vessel, not from a fillet, and should not be read as what happens to dinner.

The salmon is not the salmon the table was built on

The food side of this comparison has been moving, downward, and the table above is a snapshot of it.

Over 3,000 Scottish farmed Atlantic salmon were analysed between 2006 and 2015 (Sprague 2016, DOI 10.1038/srep21892). EPA+DHA fell from 2.74 to 1.36 g per 100 g — a halving, which matches the authors’ own phrasing that you now need double the 2006 portion to hit the same intake. Those two absolute figures reach me through a later Norwegian paper citing them, not from the original.

Two caveats, both of which cut against writing this as a collapse. First, the decline stopped: Norwegian farmed salmon has held between 1.03 and 1.30 g per 100 g from 2011 to 2020. Halved, then flat for a decade. Second, farmed still beats wild — 1.36 against 0.76 g per 100 g for wild Pacific in the same analysis, and an independent Canadian dataset found 1.50 against 0.817 (Gillies 2023, DOI 10.1155/2023/5542117). If you have been buying wild on the assumption it is the richer fish, that assumption is backwards.

There is a conflict here I cannot resolve for you. The USDA figure underneath the table above works out to about 2.15 g per 100 g of cooked farmed Atlantic salmon; Sprague’s 2015 Scottish fish measured 1.36 g raw. Some of that gap is cooking, which concentrates fat as water leaves. Some may be the age of the USDA record. Some is origin. Treat the arithmetic in this piece as the right order of magnitude and not as a measurement of the fillet in your fridge.

The line food-first writing usually leaves out

I would rather not write this paragraph, which is exactly why it goes in.

Per milligram of EPA+DHA, capsules are cheaper than fish. Not marginally — by a wide margin, in the same direction across every retail comparison I could construct. I am not publishing exact multiples, because the listings I have come from single days, some from third-party sellers rather than shelves, and prices on the same product moved between two retrievals a week apart. The direction is not in doubt. The size of it is not something I am willing to put a number on.

There is a reason this comparison is harder than it looks, and it is the more useful finding. You can compute cost per milligram for a capsule, because the label declares EPA and DHA in milligrams. You cannot do it for the tin, because canned fish labels do not declare EPA and DHA at all. Every fish figure in this article, including the ones I have just used, comes from USDA composition tables rather than from anything printed on the can. The product that is easiest to be sceptical about is the one that gives you the numbers to be sceptical with.

The reframe that keeps the cost point honest, in both directions: cost per milligram is the right denominator only if the fish is added to what you already eat. If it replaces dinner, the comparison is a tin of sardines against a capsule plus dinner, and the tin wins comfortably — it also brings protein, calcium from the bones, vitamin D and B12. In the one trial here that measured it, the salmon group gained plasma selenium the capsule groups did not — though the enzyme that selenium is supposed to support did not change, and New Zealand soil makes that finding hard to export. The capsule sells you a compound. The tin sells you a meal with the compound thrown in. A reader who leaves here thinking “capsules are cheaper, done” has been misled as badly as one who leaves thinking whole food always wins.

You cannot check the one thing that matters about a capsule

Start with a fact that deserves to be better known. In 2013, a group at the Liggins Institute in Auckland reviewed the field and reported that, to their knowledge, no human trial of omega-3 efficacy had ever reported the oxidative state of the oil it used, or compared oxidised oil against fresh (DOI 10.1155/2013/464921). Decades of trials, and the freshness of the substance under test went unrecorded.

One trial in this article is the exception, and it cuts against the capsule’s critics. Stonehouse assayed its own capsules for freshness as well as content: peroxide value 3.30 ± 0.06 meq/kg and anisidine 8.36 ± 0.67, both under the permitted maxima. Fresh, under-dosed by 40% on DHA, and still level with the fish.

Then the Auckland group tested the shelf. Of thirty-two supplements on the New Zealand market, only three met or exceeded labelled EPA+DHA and 69% contained under two-thirds of label; 83% exceeded recommended peroxide values and half exceeded TOTOX (Albert 2015, DOI 10.1038/srep07928). The work was supported by a charitable trust and declares no competing interests. It later carried a corrigendum fixing the anisidine formula and a units error — a small correction, published openly, and worth knowing about a paper this widely cited.

Two years later a second team tested forty-seven products in the same market and got close to the opposite result: 72% within the peroxide limit, 86% within anisidine, 77% within TOTOX, and 91% meeting their EPA/DHA content claims (Bannenberg 2017, DOI 10.1038/s41598-017-01470-4). Read the affiliations before you decide which one you believe. Two of the seven authors are from GOED — the industry association whose voluntary limits are the yardstick in both papers, and which the paper itself describes as an association of producers. The others are from DSM Nutritional Products, Omega Protein, and three commercial testing laboratories.

A third survey, of ten Australasian products, found every one exceeding its label for total omega-3 and all ten within the peroxide limit (Nichols 2016, PMC5133090). Its analyses were funded by the Omega-3 Centre, whose membership includes ingredient suppliers and seafood companies, and its authors declared no conflict of interest. The Auckland group published a formal Comment saying that declaration was not sustainable, since two authors were science advisors to the Centre and the third was its chair and an employee of DSM (DOI 10.3390/nu9020137). The response set out those roles, noted the advisory positions were unpaid, rejected the charge of sample bias on the grounds that the ten products covered over 80% of the Australian market, and thanked GOED’s executive director for input on the draft (DOI 10.3390/nu9060583).

I am not going to tell you which side of that is right, because the honest answer is that the market data genuinely conflict and every party to the argument has a position. What I can tell you is how to read it: when two surveys of the same shelf disagree fourfold on oxidation compliance and nearly tenfold on whether the label is true, check the author list of each before you check the abstract of either.

And check which ruler is being used. Both surveys measure against the voluntary limits set by the industry’s own trade association. Against the European and British Pharmacopoeia and Australian limits, which are less strict, the forty-seven-product survey found almost the whole shelf compliant — 98% on peroxide, 96% on TOTOX, all of them on anisidine. The argument is partly about how much rancidity is on the shelf and partly about whose yardstick counts.

Two other datasets sit between the New Zealand pair, and both are larger or older than the argument they landed in. Sixteen top-selling US liquid products: over half failed their label claims for EPA and DHA, and a quarter exceeded recommended peroxide limits (Ritter 2012, PMID 23255124). And 139 supplements in Canada, the biggest sample in this section: 17% exceeded the 5 meq O₂/kg peroxide limit (Jackowski 2015). That last figure is neither 83% nor 28%, and it comes from outside the dispute entirely.

One finding survives across all of it, and it is the one that changes a purchase. Among seventy-two US supplements sampled from 2014 to 2020, 68% of flavoured products exceeded the TOTOX limit against 13% of unflavoured ones; for peroxide value the split was 65% against 32% (Hands 2024, PMID 37712532). The authors conclude that flavourings can dramatically alter measured oxidation, to the point where flavoured products are not currently suited to the secondary-oxidation assay at all. The data came from ConsumerLab, a commercial testing business two of whose staff are co-authors; the paper reports no conflict of interest, no outside funding, and notes that ConsumerLab does not make or invest in supplements.

Put that against the shelf. The market’s answer to the burping problem is the burpless lemon capsule — and flavouring masks rancidity in your mouth and interferes with the laboratory test for it. The product engineered to be pleasant is the one that removes your only sensory check on freshness.

What you can act on: buy unflavoured. Beyond that, the Auckland survey found price, brand exclusivity, country of origin and best-before date were all poor predictors of quality, which means the folk strategy of buying expensive and refrigerating is not obviously wrong and is not verifiable by you.

And one that runs the capsule’s way. Although seafood carries varying amounts of methylmercury, omega-3 supplements have not been found to contain it, because it is removed during processing. That is the strongest single argument for the bottle in this whole file, and burying it in a food-first piece would be dishonest. It is also the older half of a benefit-risk argument that has been running since at least the 2006 JAMA review of fish intake, which concluded the cardiac benefits of modest fish consumption outweigh the contaminant risks for adults.

What the safety data supports, and what it doesn’t

Start with the numbers you can actually use, because the literature below them is contested and the regulation is not. FDA says supplement labels should not recommend more than 2 g/day of EPA+DHA, and separately concluded that supplements providing no more than 5 g/day are safe as recommended — two figures that look contradictory and are not, because one governs label claims and one governs safety assessment. EFSA puts long-term safety at around 5 g/day. No upper limit has been set. A standard capsule is 300 mg.

The main signal is atrial fibrillation, and it is genuinely contested. A meta-analysis of seven cardiovascular-outcome RCTs found AF risk raised by around 25% (Gencer 2021, PMID 34612056). What that design can show is limited: these were high-risk cardiac populations on pharmacological doses, AF was in most cases not a pre-specified primary endpoint, and ascertainment was not uniform across trials. It cannot establish incidence in a healthy adult taking a supplement.

An updated meta-analysis pooled 35 trials and 114,592 people (DOI 10.1161/CIRCEP.125.014785). Only participants at high cardiovascular risk taking more than 1,500 mg a day showed significantly increased AF: odds ratio 1.48 (95% CI 1.21–1.81), absolute risk difference 0.8%. The other three strata were null, including low-risk high-dose, which is the group closest to you and is also the smallest. Two limits matter. Every trial included ran at least twelve months, at 500 mg a day or more, in people aged fifty and over — nobody in this dataset is your age or on your dose. And the declared interests point toward softening the signal: they include the chief medical officer of a company selling omega-3 products, and, as last author, the researcher who created the omega-3 index that nearly every trial in this article uses as its outcome measure and who holds stock in the laboratory selling the test. That is the same W. S. Harris whose trade-association-funded trial opens this piece, nineteen years earlier, at both ends of the same literature. None of that makes the analysis wrong. It means you read the disclosure line before the abstract.

Countervailing evidence exists too. A 2023 meta-analysis of seventeen prospective cohorts found higher circulating marine omega-3 associated with less incident AF — in a design where the exposure is not randomised and confounding by fish intake and general health is unaddressed, and whose full citation I could not retrieve, which is itself a reason to weight it lightly.

Bleeding: the Institute of Medicine notes that 2–15 g/day might lengthen bleeding time by reducing platelet aggregation, and a 2014 review concluded omega-3s do not affect the risk of clinically significant bleeding. Data in people on warfarin come from small studies. If you take an anticoagulant, this is a conversation with whoever prescribed it, not with a website.

Prostate cancer is the one place the direction reverses, and it is worth knowing chiefly as a lesson in reading. In the SELECT case-cohort, men in the highest quartile of plasma phospholipid long-chain omega-3 had 44% higher risk of low-grade and 71% higher risk of high-grade prostate cancer than the lowest (Brasky 2013, PMID 23843441). That is a biomarker association in an observational design, and the randomised test contradicts it: VITAL, nearly 26,000 people on 1 g/day for a median 5.3 years, found no effect on prostate, breast or colorectal cancer, or on cancer mortality. NIH ODS’s summary of both is that there is no consistent relationship. When a biomarker study and a randomised trial disagree, the trial is the one that answers the question you were asking.

The side effects are the robust part. Burping 49% against 22% on placebo and unpleasant breath 39% against 18% in a 176-person trial; fishy aftertaste in 10 of 12 capsule takers against 1 of 11 fish eaters; and in the New Zealand trial, all six unprompted side-effect reports came from capsule arms. Mild, and by far the most reliable difference between the two options in this entire literature.

“Fish oil doesn’t reach your brain” — where that came from

You have probably met this claim. It has a real paper trail, and following it is the most instructive thing in this piece.

Step one is a rat study (Rodrigues 2014, PMID 24775714). Wistar rats, ten weeks, diets containing 11% or 22% canned sardines against a control, with the primary question being inflammatory markers. Among the tissues measured, the brain’s fatty acid composition was, in the paper’s words, “less affected” than the others, while adipose tissue was highly responsive.

Step two: the introduction of a 2017 mouse paper cites that study for the statement that sardines “do not appreciably increase” brain DHA in adult mammals (Sugasini 2017, DOI 10.1038/s41598-017-11766-0). Less affected than other tissues has become does not increase. That sentence sits in an introduction, as a cited assertion, not as that paper’s own result.

Step three: popular repetition drops the species. Fish doesn’t reach your brain.

A comparative statement about rats, from a study about inflammation, in three steps to a fact about you. We keep finding this shape — a number derived correctly and then applied harder than its source allows — and this is the cleanest example I have documented.

The underlying science is a live dispute, and it is worth knowing what kind. In mice, a blood-brain-barrier transporter called Mfsd2a carries DHA only in one chemical form, bound to lysophosphatidylcholine; knockouts show reduced brain DHA, neuronal loss and behavioural deficits (Nguyen 2014, DOI 10.1038/nature13241). The transporter is real in humans too: inactivating MFSD2A mutations cause microcephaly syndromes. That establishes the transporter matters. It does not establish that the chemical form of the DHA on your dinner plate matters in a healthy adult, and the whole dispute lives in that gap.

Feeding adult mice DHA in the LPC form for thirty days more than doubled brain DHA and improved performance in the probe trial of a water maze, while the same amount of free DHA did nothing for the brain (Sugasini 2017). In the same experiment, all groups learned the platform location equally during acquisition; the effect appeared only in the probe.

Then it gets contested. A rat kinetic study from a different laboratory found the reverse ordering: uptake of unesterified DHA into brain ran tenfold higher than the supposedly privileged form, and the apparent advantage on intravenous dosing was attributable to a longer plasma half-life rather than preferential transport (Chen 2015, DOI 10.1038/srep15791). A declared replication attempt in 2025 gave DHA to mice in three chemical forms for thirty days and found no significant difference in brain DHA in any region, while plasma and heart DHA rose in all treated groups — absorbed, delivered to a peripheral organ, absent from the brain (Klievik 2025, PMID 41016602). Five to six mice per group, which makes that a failure to detect rather than a demonstration of absence, and the authors say plainly they cannot explain the discrepancy.

A third laboratory, in Sherbrooke, has published on this twice, and its own results moved. In 2025 it gave genotype-specific mice LPC-bound omega-3 at nearly ten times the original dose for two to four months. Cortical DHA did not reach significance in either genotype at either timepoint — with one qualification that “no increase” would flatten: expressed as a percentage of total fatty acids, DHA did rise 22% in APOE3 mice at two months, and had not held by four. EPA rose in both genotypes, but only by four months; at two months the APOE4 animals showed no change in either fatty acid (Andriambelo 2025, PMID 39642444). In 2026, at a higher DHA dose — 10.4 mg a day of LPC-bound DHA alongside 21.5 mg of EPA — the same group reported total DHA rising in the frontal cortex of APOE3 mice but not APOE4 mice, with EPA higher across both genotypes and every compartment measured, and the choroid plexus the most responsive tissue of all (DOI 10.1016/j.jnutbio.2026.110376). Five to eight mice per group. I have read both of those at abstract level only.

So: three laboratories, a consistent EPA effect, and a DHA effect that appears and disappears with dose, duration, genotype and how the outcome is expressed. Not a settled mechanism with one dissenter, and not a debunking. Unresolved, in mice.

Now the part that makes this more than a laboratory squabble. The chemical form at the centre of the argument is a commercial ingredient — a krill-derived LPC-bound EPA/DHA launched in late 2020 and marketed on this exact transporter — and its manufacturer announced a research collaboration with the Sherbrooke group in 2021. The 2025 paper discloses this properly: the Canadian Institutes of Health Research funded the work, the company supplied the LPC oil at no cost, the company neither generated the data nor had access to the full dataset, and the authors declare no conflict of interest. Notice which way that points. The company supplied the material for a study that reported its own ingredient failing to raise brain DHA. That is the second time in this article that a commercially interested party has funded a result against its own interest, and it is the strongest available argument that these particular findings are real. The funding statement for the 2026 paper is not yet available — it is a journal pre-proof — so I cannot tell you which way its money points, and I am not assuming. Worth noting too that the laboratory behind the failed replication has no product in this at all.

The coverage is a different matter, and you can check it yourself. One trade outlet headlined the 2026 paper as the ingredient enhancing delivery of EPA and DHA to the brain, and summarised it as significant enrichment of both in brain tissues — where the paper’s own title says DHA rose in one genotype and not the other. That article also discloses that it was written with AI assistance. Another described the same work as clinical research and new clinical evidence, in subjects carrying the APOE4 gene. The subjects were mice. A genotype-restricted result in an animal model, flattened in two steps into a human claim, is the identical failure mode as the sardine sentence — running in the opposite direction, and four years faster. We ran into the same structure with krill oil, where a blood advantage was routinely reported as a brain advantage.

And the humans, briefly. The only direct measurements are PET studies. In fourteen healthy adults, whole-brain DHA consumption came to 3.8 ± 1.7 mg a day, implying a brain DHA half-life of about two and a half years (Umhau 2009); Rapoport gives 4.6 mg/day (PMID 18060754) and I cannot reconcile them. A second PET study in twenty-two adults found no significant difference by APOE genotype (PMC5364667). The Bazinet group estimates that body fat stores of DHA could supply the adult brain for fourteen to thirty-six years.

Set that against the supply side. Two servings of farmed salmon a week supply about 2,480 mg of DHA, or roughly 354 mg a day. A standard capsule is 120 mg. The organ you are buying this for consumes about 3.8. Both options overshoot it by one to two orders of magnitude — which does not prove that more is useless, but does mean that whatever limits DHA in an adult brain, the size of the delivery is not an obvious candidate.

What the cognitive trials in your age band actually found

There is exactly one positive one for you, and it is more complicated than its title.

Six months, 1.16 g DHA a day against placebo, 176 people finishing, aged 18–45, healthy, all with habitually low DHA intake — so a repletion trial, not an enhancement trial (Stonehouse 2013, PMID 23515006). Reaction time of episodic memory fell by 0.18 SD (95% CI −0.33, −0.03) — faster, because this is a reaction time. Reaction time of working memory fell by 0.36 SD (−0.58, −0.14). Then the splits: episodic memory accuracy improved in women only, by 0.28 SD (0.08, 0.48); reaction time of working memory fell in men only, by 0.60 SD (−0.95, −0.25). Processing speed was unaffected. APOE genotype made no difference. And no conclusion could be drawn on attention accuracy at all, because performance hit a ceiling — an entire domain rendered uninterpretable by test design.

The trial was funded by the Massey University Research Fund, the Neurological Foundation of New Zealand and the Oakley Mental Health Research Fund; the DHA and placebo capsules themselves were donated by two supplement companies. Academic and charitable money, an industry-donated product — worth separating, because they are not the same kind of interest.

The editorial published alongside it opens by calling the study “the first robustly designed study” of its kind and its findings potentially important, then says confidence in the conclusions is severely limited on two counts (Dangour & Allen 2013, PMID 23535110). The first is that no primary endpoint defined in advance could be identified in the article, in the trial register, or in any published protocol. In a trial reporting five cognitive domains, two sex-split subgroup findings and a genotype analysis, that is the whole methodological question. The second is that the field needs to think harder about which tests it uses.

The other trial in your band is a clean null of comparable size. A hundred and fifty-nine healthy adults aged 18–35, twelve weeks, a gram a day of either DHA-rich or EPA-rich fish oil against placebo (Jackson 2012, PMID 21864417). Serum omega-3 rose, so people took what they were given. The effects on the cognitive battery and on mood were minimal. The single finding the authors flag is that the EPA-rich oil may have reduced subjective mental fatigue during periods of high cognitive demand, which they say needs further investigation. Their conclusion is blunt: in healthy, impairment-free populations, supplementation is unlikely to produce cognitive enhancement. Right population, verified compliance, and its null is an answer rather than a puzzle.

A detail worth noticing rather than making anything of: D. O. Kennedy is a co-author of both that null and of the positive six-month trial above.

The largest evidence is in the wrong people. A meta-analysis of six RCTs and 33,496 participants, funded by the World Health Organization, found long-chain omega-3 had little or no effect on new neurocognitive illness, at a risk ratio of 0.98 (95% CI 0.87–1.10), and concludes that these supplements do not help older adults protect against cognitive decline (Brainard 2020, PMID 32305302). That is a clean null with adequate power in its own population, and it should be read as an answer for that population — not explained away. It also says nothing whatsoever about a thirty-two-year-old engineer, in either direction, and applying it to you would be the same error this article is built around. The dose-response meta-analyses that produce quotable effect sizes pool dementia patients with healthy adults; the headline attention figure in the 58-trial 2025 analysis carries a GRADE rating of low, with a confidence interval running from 0.41 to 1.54 (DOI 10.1038/s41598-025-16129-8).

The gap that matters most: no trial has ever randomised healthy adults to fish versus capsules with a cognitive endpoint. Every cognitive result in this literature is capsules against placebo. Nobody has tested the question you are actually asking in the aisle. If you want the general version of why these trials so often come back empty, we wrote it up separately — it is the most transferable thing on this site.

One more piece of context, because it reframes the scale of everything above. US adults get about 90 mg of EPA+DHA a day from food, against an EFSA adequate intake of 250 mg. Supplements add, on population average, about 30 mg a day — a tenth of one capsule. And NIH ODS states there is no known concentration of EPA or DHA below which functional endpoints are impaired. There is no deficiency threshold for the thing you are topping up.

The regulator’s own hedge, on the label of the thing you are holding, is that the research is “supportive but not conclusive.”

The bet

The thing you are buying this for has no demonstrated payoff in people like you. The things you get incidentally — a meal, a cost, a mercury exposure, a burp — are documented and certain. So make the decision entirely on the certain terms.

Take the capsule if the realistic alternative is not eating fish at all — which is exactly who the best trial here recruited. It is cheap, precise, mercury-free, and it will move your blood number as reliably as fish will. Accept that you cannot verify its freshness, that flavoured products exceeded the secondary-oxidation limit at roughly five times the rate of unflavoured ones and the peroxide limit at twice the rate, that label under-delivery is common, and that around half of people get the burping.

Eat the fish if it displaces a meal you were going to eat anyway. Accept that it costs more per milligram, that the species choice matters roughly tenfold and the tin you probably buy is the wrong one, and that the farmed salmon underneath the composition tables is not the salmon of twenty years ago.

What should not enter the decision either way is the expectation that it changes how well you think this quarter. In your age band there is one positive trial with sex-split results and an editorial noting no pre-specified primary endpoint, and one null of comparable size. That is the whole file.

Four things to do on Monday, in descending order of size:

1. Change the tin. Canned pink salmon or sardines instead of light tuna is a five- to sixfold move in the same aisle at the same money, and it is larger than the entire fish-versus-capsule question.
2. If you take capsules, take them with the fattiest meal of your day, not with a coffee.
3. Buy unflavoured.
4. Don’t deep-fry the fish. Everything else — baking, steaming, pan-frying, the microwave — appears interchangeable.

The asymmetry is this: you lose nothing by ignoring the brain claim, and you lose real money or real dinners by organising the decision around it.

About this article

Written by Leah Elish. Leah covers food-first nutrition — what a nutrient does, how much of it is in a meal, and what the capsule adds to that. Not a dietitian or clinician — traces nutritional claims back to the studies that supposedly support them.

Medical review: None. NeuriFuel does not currently have a licensed clinician on the editorial team, and this article has not been medically reviewed. We state this rather than implying an authority we do not have. See our About page for our full methodology.

Sources: 57 — fifteen human trials or imaging studies, seven animal studies, four meta-analyses, twelve compositional, product-analysis or observational studies, seven reviews, editorials, comments, responses or corrections, eight guidance and reference documents, and four trade-press and company sources used only to document marketing language and never for study results. Three sources were read in full text (Stonehouse 2011, Sharman 2025 and the four-sentence 2016 corrigendum to the New Zealand supplement survey); most were read at abstract or index-record level, and several are known only through descriptions in later papers, which is marked in the text wherever it affects what can be claimed. Values that could not be retrieved are marked as such rather than estimated. Nine of the entries below carry an explicit note that funding could not be retrieved; many others carry no funding line at all, which means silence in the list should not be read as a clean disclosure. Where two sources give conflicting values — the ninefold versus threefold description of one trial, the two brain-DHA consumption figures, the two surveys of the New Zealand supplement market — both are reported and neither is chosen. One paper relied on here, the 2015 New Zealand supplement survey, carries a published corrigendum, noted in the text. One paper cited in the safety section, a 2023 meta-analysis of seventeen cohorts, has no entry below because its full citation could not be retrieved.

Corrections: Found an error? Write to hello@neurifuel.com with a source and we will fix it and log the correction.

Last updated: 18 August 2026

References

Human trials and imaging studies

1. Harris WS, Pottala JV, Sands SA, Jones PG (2007). Am J Clin Nutr 86(6):1621–1625. PMID 18065578. — n = 11 fish, 12 capsules. Part-funded by a Fisheries Scholarship grant from the National Fisheries Institute, a US seafood trade association, and by the Saint Luke’s Hospital Foundation. WSH a scientific advisor to Monsanto and Reliant Pharmaceuticals; SAS employed by OmegaMetrix LLC.
2. Stonehouse W, Pauga MR, Kruger R, Thomson CD, Wong M, Kruger MC (2011). Br J Nutr 106(8):1231–1239. DOI 10.1017/S000711451100153X. — Salmon supplied by The New Zealand King Salmon Company Limited.
3. Visioli F, Risé P, Barassi MC, Marangoni F, Galli C (2003). Lipids 38(4):415–418. PMID 12848287. — Pilot, 4–8 volunteers per group, retrospective, non-randomised, intakes unmatched. Named funder not retrieved. Known here only through descriptions by later authors.
4. Elvevoll EO, Barstad H, Breimo ES, et al. (2006). Lipids 41:1109–1114. PMID 17269556. — Population, N, effect sizes and funding not retrieved.
5. Arterburn LM, Oken HA, Bailey Hall E, Hamersley J, Kuratko CN, Hoffman JP (2008). J Am Diet Assoc 108:1204–1209. PMID 18589030. — Four of six authors were Martek Biosciences personnel.
6. Sharman A, Zhou R, Pugh J, Close G, Fisk HL, Calder PC, Yaqoob P (2025). Br J Nutr 133(7):934–944. PMID 40114588. — Funded by a grant to Liverpool John Moores University from BBC “Trust Me I’m A Doctor”; supplements donated by Wiley Companies, meals by Soulmate Food; study designed together with the funders; no conflicts declared. ISRCTN13031606.
7. Lawson LD, Hughes BG (1988). Biochem Biophys Res Commun 152(1):328–335. PMID 3358766. — n = 8; funding not retrieved.
8. Lawson LD, Hughes BG (1988). Biochem Biophys Res Commun 156(2):960–963. PMID 2847723. — Sample size and funding not retrieved.
9. Nordøy A, et al. (1991). Am J Clin Nutr — full citation and funding not retrieved.
10. Dyerberg J, et al. (2010). Bioavailability of marine n-3 formulations — full citation and funding not retrieved.
11. Umhau JC, et al. (2009). J Lipid Res 50:1259–1268 — volume, pages and author list unverified.
12. Rapoport SI (2008). Brain metabolism of long-chain polyunsaturated fatty acids. PMID 18060754. — Source of the 4.6 mg/day figure conflicting with ref. 11.
13. Yassine HN, et al. PET study of brain DHA incorporation by APOE genotype, 22 adults. PMC5364667. — Author list and full citation not retrieved.
14. Stonehouse W, Conlon CA, Podd J, Hill SR, Minihane AM, Haskell C, Kennedy D (2013). Am J Clin Nutr 97(5):1134–1143. PMID 23515006. — Supported by the Massey University Research Fund, the Neurological Foundation of New Zealand and the Oakley Mental Health Research Fund; supplements supplied by Efamol Ltd and Health & Herbs International Ltd.
15. Jackson PA, Deary ME, Reay JL, Scholey AB, Kennedy DO (2012). Br J Nutr 107(8):1232–1243. PMID 21864417. — Funding not retrieved.

Animal studies

16. Nguyen LN, Ma D, Shui G, et al. (2014). Nature 509:503–506. PMID 24828044.
17. Sugasini D, Thomas R, Yalagala PCR, Tai LM, Subbaiah PV (2017). Sci Rep 7:11263. — Funding not retrieved.
18. Rodrigues PO, Martins SV, Lopes PA, et al. (2014). Br J Nutr 112(3):309–319. PMID 24775714.
19. Chen CT, Kitson AP, Hopperton KE, et al. (2015). Sci Rep 5:15791.
20. Klievik BJ, Fu Y, Tyrrell AD, Chen CT, Metherel AH, Bazinet RP (2025). J Lipid Res 100913. PMID 41016602. — n = 5–6 per group. Funding not retrieved.
21. Andriambelo B, Vachon A, Dansereau M-A, Laurent B, Plourde M (2025). Prostaglandins Leukot Essent Fatty Acids 204:102661. PMID 39642444. — Funded by the Canadian Institutes of Health Research; LPC n-3 oil supplied at no cost by Aker BioMarine Human Ingredients AS, which did not generate the data or have access to it; no conflict declared.
22. Andriambelo B, Vachon A, Dansereau M-A, Dumais É, Flamand N, Laurent B, Plourde M (2026). J Nutr Biochem 155:110376. — Journal pre-proof; funding statement not available.

Meta-analyses

23. Gencer B, Djousse L, Al-Ramady OT, Cook NR, Manson JE, Albert CM (2021). Circulation 144(25):1981–1990. PMID 34612056.
24. Abuknesha NR, O’Keefe JH, Qian F, Tintle NL, et al., Harris WS (2026). Circ Arrhythm Electrophysiol e014785. DOI 10.1161/CIRCEP.125.014785. — Declared interests include the chief medical officer of a company selling omega-3 products and stock in an omega-3 blood-testing laboratory. Participants aged 50 and over.
25. Brainard JS, Jimoh OF, Deane KHO, et al.; PUFAH group (2020). J Am Med Dir Assoc 21(10):1439–1450.e21. PMID 32305302. — Supported by the World Health Organization, Grant HQNHD1612458.
26. Systematic review and dose-response meta-analysis of omega-3 supplementation on cognitive function (2025). Sci Rep. DOI 10.1038/s41598-025-16129-8; PMC12368174. — 58 RCTs. Funding not retrieved.

Compositional, product-analysis and observational studies

27. Sprague M, Dick JR, Tocher DR (2016). Sci Rep 6:21892. PMID 26899924.
28. Nøstbakken OJ, et al. (2021). Food Chem — full citation not retrieved.
29. Gillies M, et al. (2023). Aquaculture Research. DOI 10.1155/2023/5542117.
30. Larsen D, Quek SY, Eyres L (2010). Food Chem 119:785–790. — Known here through a later paper’s description.
31. Al-Saghir S, et al. (2004). — Known here only through the literature review in ref. 30.
32. Albert BB, Derraik JGB, Cameron-Smith D, Hofman PL, Tumanov S, Villas-Boas SG, Garg ML, Cutfield WS (2015). Sci Rep 5:7928. DOI 10.1038/srep07928. — Supported by The Douglas Charitable Trust; no competing financial interests declared. See corrigendum, ref. 41.
33. Ritter JCS, et al. (2012). J Sci Food Agric. PMID 23255124.
34. Jackowski SA, et al. (2015). — full citation not retrieved.
35. Bannenberg G, Mallon C, Edwards H, Yeadon D, Yan K, Johnson H, Ismail A (2017). Sci Rep 7:1488. PMID 28469193. — Two of seven authors from GOED, the omega-3 industry association that sets the limits applied; others from DSM Nutritional Products, Omega Protein and three commercial testing laboratories.
36. Nichols PD, Dogan L, Sinclair A (2016). Nutrients 8(11):703; PMC5133090. — Analyses funded by the Omega-3 Centre, whose membership includes omega-3 ingredient suppliers and seafood companies. No conflict of interest declared.
37. Hands JM, Anderson ML, Cooperman T, Frame LA (2024). J Diet Suppl 21(2):195–206. PMID 37712532. — Two of four authors are of ConsumerLab.com. No conflict reported; no outside financial assistance; data provided by ConsumerLab LLC.
38. Brasky TM, et al. (2013). J Natl Cancer Inst 105:1132–1141. PMID 23843441.

Reviews, editorials, comments, responses and corrections

39. Albert BB, Cameron-Smith D, Hofman PL, Cutfield WS (2013). BioMed Res Int 2013:464921. PMID 23738326. — Funding not retrieved.
40. Mozaffarian D, Rimm EB (2006). JAMA 296(15):1885–1899. PMID 17047219. — Known here through secondary summaries.
41. Albert BB, et al. (2016). Corrigendum. Sci Rep 6:35092.
42. Albert BB, Derraik JGB, Garg ML, Cameron-Smith D, Cutfield WS (2017). Nutrients 9(2):137. PMID 28216562.
43. Nichols PD, Dogan L, Sinclair A (2017). Nutrients 9(6):583. — Two authors unpaid science advisors to the Omega-3 Centre; the third its unpaid chair and an employee of DSM. Acknowledges input from GOED on the draft.
44. Dangour AD, Allen E (2013). Am J Clin Nutr 97(5):909–910. PMID 23535110.
45. Bazinet group estimate of adipose DHA stores (20–50 g) sufficient to supply the adult brain for 14–36 years. — Full citation not retrieved.

Guidance and reference documents

46. NIH Office of Dietary Supplements. Omega-3 Fatty Acids: Fact Sheet for Health Professionals. — Source for the VITAL, Institute of Medicine and bleeding-review summaries used here.
47. USDA FoodData Central, EPA and DHA content of seafood, as tabulated by NIH ODS. Per-100 g conversions in this article are ours.
48. Siscovick DS, Barringer TA, Fretts AM, et al. (2017). Omega-3 polyunsaturated fatty acid (fish oil) supplementation and the prevention of clinical cardiovascular disease: a science advisory from the American Heart Association. Circulation 135:e867–e884. DOI 10.1161/CIR.0000000000000482.
49. Rimm EB, Appel LJ, Chiuve SE, Djoussé L, Engler MB, Kris-Etherton PM, Mozaffarian D, Siscovick DS, Lichtenstein AH (2018). Seafood long-chain n-3 polyunsaturated fatty acids and cardiovascular disease: a science advisory from the American Heart Association. Circulation 138:e35–e47. DOI 10.1161/CIR.0000000000000574; PMID 29773586.
50. Skulas-Ray AC, et al. (2019). Omega-3 fatty acids for the management of hypertriglyceridemia: a science advisory from the American Heart Association. Circulation. DOI 10.1161/CIR.0000000000000709.
51. US Food and Drug Administration. Qualified health claim for EPA and DHA and coronary heart disease (2004).
52. US Food and Drug Administration. Labelling and safety statements on EPA+DHA intake.
53. European Food Safety Authority. Adequate intake for EPA and DHA in adults.

Trade-press and company sources (used only to document marketing language, never for study results)

54. Nutritional Outlook, 17 April 2026. — Discloses AI assistance in its creation.
55. NutritionInsight, 14 April 2026.
56. Aker BioMarine, Lysoveta product pages, 2020–2026.
57. Aker BioMarine research collaboration announcement, 17 February 2021.