Vitamin K2 (MK-7)

Cartilage contains vitamin-K-dependent proteins, and the important one is matrix Gla protein. It only inhibits mineralisation once vitamin K has carboxylated it, and mice that lack it entirely calcify their arteries and cartilage and die of vascular rupture within about two months. Inappropriate cartilage calcification is a recognised feature of osteoarthritis, so the hypothesis writes itself: keep vitamin K status adequate, keep matrix Gla protein switched on, keep cartilage un-mineralised. Two more vitamin-K-dependent proteins turn up in the same tissue — Gla-rich protein, which inhibits calcification when carboxylated and damps inflammation whether carboxylated or not, and Gas6, which recent work makes the route by which vitamin K blocks a form of chondrocyte cell death.

Why MK-7 rather than MK-4 or K1

Vitamin K2 is a family, and the members behave differently enough that the distinction decides what a capsule does.

Against K1, MK-7 is absorbed just as well and peaks in serum at the same four hours, but it clears far more slowly. That gives much steadier blood levels and seven- to eightfold higher accumulation during ongoing intake, and it carboxylates osteocalcin more completely. Against MK-4, the gap is starker: a single 420 µg dose of MK-4 was undetectable in serum at every sampling point in healthy women, while MK-7 was well absorbed and accumulated on 60 µg a day. At the doses printed on supplement labels, MK-4 does not reach the circulation and MK-7 does.

That matters for reading the older literature. A large share of vitamin K2's public reputation comes from Japanese osteoporosis trials that used MK-4 at 45 mg a day — a different molecule at roughly 225 times a 200 µg MK-7 dose, and a pharmaceutical rather than nutritional intervention. Those results do not scale down.

The dose-response for MK-7 itself has been mapped. Across seven arms from 10 to 360 µg a day in healthy adults, doses at and above the recommended daily allowance increased carboxylation of both osteocalcin and matrix Gla protein while lower doses did not reliably do so. In dialysis patients — the group with the most severe functional deficiency — the inactive form of matrix Gla protein fell dose-dependently, with 93% of patients responding at 360 µg a day, and in a 200-patient dose-finding study the marker dropped 17%, 33% and 46% across three ascending doses. Common commercial doses of 90–200 µg sit above the threshold where the marker moves and below the pharmacological range used in kidney disease.

What MK-7 is taken for: arteries and bone

The multi-year MK-7 trials measured cardiovascular and skeletal endpoints, and the results are genuinely mixed rather than uniformly positive.

Two hundred and forty-four healthy postmenopausal women took 180 µg a day for three years. Inactive matrix Gla protein fell by half against placebo, carotid-femoral pulse wave velocity and the stiffness index improved across the whole group, and local carotid measures improved most in the women whose arteries were stiffest to begin with. The same cohort's bone arm found slower decline in bone mineral density at the lumbar spine and femoral neck, better calculated femoral-neck strength, and less loss of vertebral height — density and strength indices, not fractures.

On coronary calcium the 2026 result is positive and carefully hedged. In 180 patients with symptomatic coronary disease, calcium scores rose in both arms over two years but less on 360 µg of MK-7 — a median rise to 184 Agatston units against 214 on placebo. The authors note the increase tracked soft plaque becoming partly calcified, which leaves open what slowing it means for plaque stability.

Then the nulls. The largest trial in this space gave 720 µg of MK-7 with vitamin D to 365 men with advanced aortic valve calcification for two years: the difference in valve calcification was 17 Agatston units, with nothing on valve area, jet velocity, or aortic and coronary calcification — even though the inactive matrix Gla protein marker fell sharply. In type 2 diabetes, 360 µg for six months nudged femoral calcification activity on PET in the wrong direction, again while the marker fell as designed. In vitamin-K- deficient kidney transplant recipients the primary endpoint was flat and only arterial stiffness separated.

The pattern across those trials is worth stating plainly, because it is the strongest caution in the file: MK-7 reliably normalises the carboxylation biomarker, and the biomarker does not reliably carry the outcome.

At the level of diet rather than supplements, the Rotterdam Study found the top tertile of menaquinone intake had less than half the coronary heart disease mortality and severe aortic calcification of the bottom tertile, while phylloquinone intake was related to none of it — the observation that started the menaquinone story, and an observation about cheese and fermented food as much as about a molecule.

The joint evidence

The observational record is unusually consistent. In Framingham Offspring, higher plasma vitamin K tracked lower prevalence of radiographic hand osteoarthritis and knee osteophytes. In MOST, subclinical deficiency predicted incident radiographic knee osteoarthritis over 30 months and incident MRI cartilage lesions. In Health ABC, very low plasma vitamin K predicted progression of cartilage and meniscus damage over three years. In the Osteoarthritis Initiative, lower dietary vitamin K tracked faster loss of medial joint space width in a dose-response pattern. In an Australian cohort of people with symptomatic knee osteoarthritis, higher intake predicted greater symptom improvement over two years, though MRI structure did not follow. A clinic case-control found serum vitamin K lower in knee osteoarthritis patients than controls.

The most striking human evidence comes from the opposite direction — switching the pathway off. Vitamin K antagonist anticoagulants inactivate matrix Gla protein as a side effect of inactivating clotting factors. In the Rotterdam Study, users of one such drug had 2.50 times the risk of osteoarthritis incidence and progression, 2.34 at the knee and 2.74 at the hip; users who carried both a high-expression clotting-enzyme haplotype and the matrix Gla protein risk allele reached 4.18, while that same genotype carried no risk at all in non-users. In a UK primary-care database, among people all anticoagulated for atrial fibrillation, warfarin users had 1.59 times the knee or hip replacement risk of those on direct oral anticoagulants, rising with duration of use. That comparison matters because both groups needed anticoagulating, which strips out the crudest confounding.

The one randomised trial with a joint endpoint supplemented K1, not K2, as an ancillary to a bone trial in 378 people over three years. It was null on radiographic hand osteoarthritis. Participants insufficient at baseline who reached sufficiency showed about 47% less joint-space narrowing, which is a post-hoc subgroup and hypothesis-generating only.

The findings that cut against it

Mendelian randomization, the cleanest available test of causality, came back null. Using the largest osteoarthritis genome-wide study to date — up to 826,690 people — genetically predicted vitamin K showed no association with osteoarthritis overall, or at knee, hip, spine or hand, or with joint replacement, across every sensitivity analysis. The genetic instrument is weak, derived from a study of only 2,138 people, so a modest true effect could hide inside those intervals. It is still a null.

The only human study to measure menaquinones directly against an osteoarthritis endpoint also found nothing: faecal menaquinone profiles did not differ between obese participants with and without hand and knee osteoarthritis. Every other human joint study measured phylloquinone, which means the case for taking K2 specifically rests on pharmacokinetics rather than on outcomes.

The cohorts confound vitamin K status with overall diet quality and health behaviour, and the vitamin-K-antagonist studies test blocking the pathway rather than supplementing it, in patients who differ from everyone else in ways beyond their prescription.

The preclinical work adds a complication of its own. Vitamin K2 protected cartilage in a rat surgical osteoarthritis model, and vitamin K came out of an FDA-library screen as the strongest anti-ferroptotic compound in chondrocytes, reducing disease in mice — but both worked through antioxidant and cell-death pathways rather than through calcification, and the mouse study injected into the joint. A third study found MK-4 with calcitriol protecting cartilage indirectly, by suppressing osteoclasts in the bone beneath it. Real effects, in a different mechanism from the one the supplement is usually sold on.

Dose & practical notes

No joint trial establishes a dose. The multi-year MK-7 outcome trials used 180 µg for arterial stiffness and bone and 360 µg for coronary calcium, and the carboxylation dose-response puts the reliable threshold at around 180 µg. Common commercial doses of 90–200 µg sit inside that band. MK-7 is fat-soluble and the trials dosed it with meals. The one joint signal that exists came from repleting people who were insufficient rather than from supra-nutritional dosing, which makes this a nutrient-status story rather than a pharmacology one.

Safety

Vitamin K has low toxicity at nutritional and common supplemental doses, with no adverse effects reported across the two- and three-year MK-7 trials and no effect on thrombin generation anywhere from 10 to 360 µg a day in healthy subjects.

The warfarin interaction is the clinically important issue and it is quantified. In healthy volunteers anticoagulated to a target INR of 2.0, MK-7 at 45 µg a day cut mean INR and uncarboxylated clotting factor II by about 40%, and even 10 µg a day — a twentieth of a 200 µg capsule — produced a clinically relevant INR fall in at least 40% of subjects. The pharmacokinetic work puts the interference threshold around 50 µg a day. Anyone taking warfarin or a related vitamin K antagonist should not add K1 or K2 without medical supervision.

The irony is coherent and worth stating without overreach: those same drugs inactivate matrix Gla protein, and their users carry more osteoarthritis progression and more joint replacements. That is a reason for a conversation with a prescriber about which anticoagulant, not a reason to self-supplement while anticoagulated.

What would move this entry

The mechanism is demonstrated down to the knockout mouse, six cohorts point the same way, and the vitamin-K-antagonist data are as close to a human experiment as observational work gets. Set against that are a null randomised trial, a null causal-genetics analysis, and a run of vascular trials in which the biomarker moved and the endpoint did not. The study that would settle it is a straightforward one: MK-7 at a label dose, in adults who are genuinely vitamin-K insufficient, with knee MRI over several years. The largest MK-7 trial now running is measuring coronary calcium, bone, lungs and metabolism in 450 people over three years — and no joint.