Omega-3s & fish oil

Omega-3s have a mechanism that resolves into named molecules and a trial literature that argues with itself. For years the summary was straightforward — good biology, null trials. That is no longer accurate, and the way it stopped being accurate is more interesting than either version.

The mechanism is real

In cartilage explants and chondrocytes, n-3 fatty acids incorporated into cell membranes dose-dependently reduce aggrecanase activity and switch off expression of IL-1α, TNF-α and COX-2 — the core catabolic machinery of cartilage breakdown. The effect is specific to n-3s; n-6 and saturated fats do not do it. Omega-3s also compete with arachidonic acid, the substrate for pro-inflammatory eicosanoids, and in humans higher arachidonic acid levels track with knee synovitis.

The cleanest version of the hypothesis is a mouse that supplements itself. The fat-1 transgenic mouse converts n-6 fatty acids to n-3 endogenously, so its tissues carry a different fatty-acid composition with nothing fed to it. After meniscal resection those mice had significantly less cartilage destruction and osteophyte formation (p < 0.01), lower MMP-13 and ADAMTS-5 protein in articular cartilage (p < 0.01), and less chondrocyte and matrix loss — through mTORC1 downregulation and increased chondrocyte autophagy. That is a study about tissue omega-3 status, not about a capsule, and it is the reason the hypothesis has survived its trials.

The two krill oil trials

In 2022, 235 adults aged 40 to 65 with mild-to-moderate knee osteoarthritis took 4 g/day of krill oil — 0.60 g EPA, 0.28 g DHA, 0.45 mg astaxanthin — or a vegetable-oil placebo for six months, across multiple centres and double blind. Their Omega-3 Index rose from 6.0% to 8.9% while placebo stayed flat. Knee pain improved in both arms and more on krill oil: adjusted between-group difference −5.18 on a 0–100 scale (95% CI −10.0 to −0.32, p = 0.04), with stiffness (−6.45) and physical function (−4.67) moving the same way.

In 2024, 262 adults with knee osteoarthritis took 2 g/day of krill oil or placebo for 24 weeks — and this trial deliberately selected people with effusion-synovitis on MRI, the inflammatory phenotype most expected to respond. Pain improved identically in both groups, with a between-group difference of about a third of a point on a 100-point scale and a confidence interval straddling zero. WOMAC pain, function, hand pain and back pain were null alongside it.

One of that trial's sixty secondary outcomes did separate, and it went to placebo. Effusion-synovitis volume — the inflammation people had been recruited on — rose by 0.81 mL on krill oil and fell by 0.94 mL on placebo, a difference of 1.75 mL (p = 0.01). The authors call the result unexpected and say that at sixty outcomes it may be chance, which is the right thing to say about it. It is still the only endpoint in the trial that moved.

Two well-conducted trials of the same supplement, two years apart, in the same disease, disagreeing.

The entry criterion that may explain it

The 2022 trial required participants to be consuming less than 0.5 grams a day of long-chain omega-3 to be eligible. It was, by construction, a repletion study: it found people who were short and made them not short.

If that is what the difference is, the question changes shape entirely. It stops being "do omega-3s treat osteoarthritis" and becomes "does correcting an omega-3 deficit help a joint" — which would apply to a defined minority and not to everyone, and which would explain why trials that did not screen on intake keep coming back flat. The fat-1 mouse points the same way, because it is a status experiment rather than a dosing one.

This is not the site reading something into two trials. The 2024 authors reach for the same explanation themselves: their participants started at a higher Omega-3 Index than the 2022 trial's, finished at a lower one, and were never asked to keep their dietary omega-3 down, which the 2022 trial required. They put a lower dose beside it as the other candidate. Either way the variable is status rather than the supplement.

Nobody has tested this. Every modern trial measures an Omega-3 Index; none has stratified on it.

Two other observations sit against the positive trial. Its inflammatory markers did not move at all, leaving the proposed mechanism unsupported inside the trial that found the effect. And five points on a hundred-point scale, with a confidence interval reaching −0.32, is a small effect measured precisely rather than a meaningful one measured loosely.

The rest of the human record

The next-largest trial compared high-dose fish oil (4.5 g/day omega-3) against low-dose (0.45 g/day) for two years in 202 knee-OA patients. High dose was no better — the low-dose arm actually improved more on pain and function, and MRI cartilage volume loss did not differ. With no placebo arm no efficacy conclusion is possible in either direction, and the low-dose-wins pattern has the shape of regression to the mean. The same trial's bone-density arm was equally null: no difference in lumbar spine or femoral neck BMD after two years.

A meta-analysis of 42 marine-oil trials found a small overall pain effect driven almost entirely by rheumatoid arthritis; in the five osteoarthritis trials the effect was non-significant, and the authors graded that subgroup very low. RA is an autoimmune inflammatory disease and its positive signal does not transfer to OA cartilage. A 2022 pooled analysis of nutritional supplements found only two omega-3 trials to work with, and in them WOMAC function improved while pain did not — the endpoint people take fish oil for is the one that did not move.

And in 249 women with aromatase-inhibitor joint pain, 3.3 g/day of omega-3 took worst pain from 7.06 to 4.83 over 24 weeks. Placebo took it from 6.88 to 5.07, and the two never separated at any timepoint (p = 0.58). Triglycerides fell only on omega-3, confirming the capsules were being taken and absorbed. That trial is worth keeping in view for the size of the shared improvement rather than for the null: about a third of the pain went in both arms, joint pain is an endpoint with an enormous placebo response, and any trial in this area is measuring against it.

One positive human record does not concern joints at all. A Japanese trial reported knee-pain improvement from 149 mg of EPA — a twentieth of the OA trial doses — combined with l-serine, from authors employed by the company selling it. Nothing in that design attributes the result to EPA.

The cohort signal, and where it points

The one genuinely encouraging human cartilage datum is observational. In 472 adults from the MOST cohort, higher plasma omega-3 and DHA levels were associated with less patellofemoral cartilage loss — though not tibiofemoral — while n-6 arachidonic acid was associated with synovitis. It carries every limitation of a cross-sectional cohort, and it is the strongest cartilage-specific human signal for any nutrient here.

Whole fish behaves differently from capsules in the observational data, and that divergence has its own entry: in 35,331 adults linked to a national joint-replacement registry, fish intake predicted neither hip nor knee replacement.

What the dogs say

Veterinary omega-3 trials read more positively than human ones and are frequently cited as if that carried over. In 127 dogs across 18 clinics, a food with 31 times more omega-3 improved owner-reported ability to rise, play and walk over six months. But the two diets were physically different products, every clinical outcome was an owner questionnaire, and several authors worked for the pet-food manufacturer.

The un-commissioned trial is the useful one. In 77 dogs randomized to fish oil or corn oil for 16 weeks, force-plate measurements, a validated pain index and rescue NSAID use showed no significant difference between groups — while within the fish-oil arm, change from baseline looked good. Owners of fish-oil dogs were also significantly better at guessing which group they were in.

Fish oil is not a blindable intervention in any species. That is worth carrying back into the human literature, where within-group improvement on an unblindable supplement is exactly the finding that keeps getting quoted.

Where that leaves you

Safety is not the obstacle: adverse events matched placebo in the krill trials, and the known considerations are mild GI effects, fishy burps, dose-dependent effects on bleeding time at high intakes, and caution alongside anticoagulants.

The defensible reading has moved. It is no longer that marine oil does nothing; it is that it may do a little, for people who were short of it, on symptoms rather than structure. The trial that would settle that — marine oil stratified by baseline Omega-3 Index, with an imaging endpoint — is now the most obviously missing study in this section, and every trial run in the last decade has already collected the variable it would stratify on.