Drink 25 grams of fish gelatin hydrolysate and, about an hour later, an intact two-amino-acid fragment of it is circulating in your blood. Yoshifumi Taga and colleagues at the Nippi Research Institute of Biomatrix measured precisely that, quantifying hydroxyproline and thirteen hydroxyproline-containing peptides in human blood by mass spectrometry against a stable isotope-labelled collagen standard. The dipeptide prolyl-hydroxyproline reached an area under the plasma concentration-time curve of 163 nmol/mL per hour.

Which makes the most popular argument against collagen supplements wrong on the biochemistry. That argument says digestion dismantles collagen into ordinary amino acids, so the building-block story collapses before anything is absorbed. It does not. Fragments of collagen get in, and they get in whole.

They also do not go where the label implies. The claim on the tub and the standard debunking of it are both wrong, and the gap between them is where the real question sits: collagen fragments survive digestion and enter the bloodstream, but nothing then routes them to the skeleton, and no trial in this literature has counted a single fracture. What the human evidence can support is a narrower and more interesting proposition — that collagen-derived peptides act as a signal rather than as raw material. That proposition currently rests on one twelve-month randomised trial, a scattering of small short ones, and a mouse cell line.

The fragments survive digestion, and that part of the label is defensible

Taga's group was not alone. Yasutaka Shigemura and colleagues at Tokyo Kasei University gave healthy volunteers three escalating doses of collagen hydrolysate — 30.8, 153.8 and 384.6 mg per kilogram of body weight — and watched plasma hydroxyproline-containing peptides rise in step: peak concentrations of 6.43, 20.17 and 32.84 nmol/mL. The absorption ceiling was not reached even at the top dose. Four volunteers, so read those figures as a demonstration of a mechanism rather than as population norms.

Shigeru Yamamoto's team at Jellice went further and followed the peptides out the other end. After volunteers ingested a collagen preparation rich in Gly-Pro-Hyp tripeptides, both Gly-Pro-Hyp and Pro-Hyp appeared in plasma and were then excreted in the urine still in peptide form — meaning the fragments were stable enough in the body to be filtered by the kidney without ever being broken down.

There is a known door for this. The oligopeptide transporter PepT1, sitting in the intestinal wall, exists specifically to carry di- and tripeptides across intact rather than dismantling them first. It is such a reliable route that pharmaceutical chemists deliberately disguise poorly absorbed drugs as peptides to smuggle them through it. Collagen hydrolysate is simply a very large supply of exactly the shape that transporter was built for.

Surviving digestion is not the same as being delivered

The building-block story requires a second step that nobody has demonstrated: that these circulating fragments are carried to bone and fitted into its matrix. The one tracer study available points somewhere else. Mari Watanabe-Kamiyama and colleagues at Tohoku University gave Wistar rats collagen hydrolysate radiolabelled either on proline or on the tripeptide glycyl-prolyl-hydroxyproline. Plasma radioactivity rose quickly, the tripeptide persisted in plasma and accumulated in the kidney — and radioactivity stayed high in the skin for fourteen days after a single dose. This is a rodent study, and radioactivity in a tissue is not proof of matrix incorporation. But the tissue it points at is skin, not bone.

What the human data do show is a body making more of its own collagen after a dose. Greg Shaw and colleagues at the Australian Institute of Sport ran a randomised double-blind crossover in eight healthy men: 15 grams of vitamin C-enriched gelatin taken an hour before six minutes of rope-skipping doubled circulating amino-terminal propeptide of collagen I, a marker of type I collagen synthesis. Eight men, one marker, one exercise stimulus. But the shape of the finding matters more than its size: it is consistent with collagen acting as a trigger for synthesis, and inconsistent with it arriving as prefabricated matrix.

If collagen does something for bone, it does it as a signal, and that case is built on a mouse cell line

Yoshifumi Kimira's group at Josai University added Pro-Hyp to MC3T3-E1 mouse osteoblastic cells and found alkaline phosphatase activity rose and the osteogenic genes Runx2, Osterix and Col1a1 were upregulated. Two details in that result are easy to skip past. Pro-Hyp did not increase cell proliferation, and it did not increase matrix mineralisation — the two things a building block would be expected to do. A later paper from the same group worked out the mechanism: Pro-Hyp binds the transcription factor Foxg1 directly, changing its conformation and releasing it from Runx2.

That is a signalling story, and its authors know it. It is also a cultured mouse cell line, two long steps from a postmenopausal woman, and the peptide division of a gelatin manufacturer co-authored both papers. Worth noting that even collagen's laboratory advocates are not making the building-block argument. The marketing is a claim the mechanistic literature abandoned some time ago.

One year, 131 women, and a tenth of a standard deviation

The single best piece of human evidence is a randomised, double-blind, placebo-controlled trial run by Daniel König's group at the University of Freiburg. They enrolled 131 postmenopausal women with age-related reductions in bone mineral density (baseline spine T-score −2.4 ± 0.6) and gave them 5 grams of specific collagen peptides or placebo daily for twelve months. A hundred and two completed; all 131 were analysed by intention to treat.

Spine T-score changed by +0.1 ± 0.26 in the peptide group against −0.03 ± 0.18 in controls (ANCOVA p = 0.030). At the femoral neck, +0.09 ± 0.24 against −0.01 ± 0.19 (p = 0.003). The bone formation marker P1NP rose in the peptide group (p = 0.007), while the resorption marker CTX-1 rose in the control group (p = 0.011).

A T-score is denominated in standard deviations, so the between-group difference after a year is somewhere around a tenth of one standard deviation. Look at the spread rather than the average and the picture gets more honest still: the standard deviation of the change is two to three times the size of the mean change, which means a substantial share of women who took the peptides did not move up at all.

A four-year follow-up exists and is much weaker than its conclusion sounds. It was open-label with no control group, and 31 of the original 131 women completed it. The disclosure statement notes that one author is a co-inventor of patents covering the use of collagen peptides and has given lectures partly supported by a gelatin manufacturer. An uncontrolled follow-up of the people who stayed in is the weakest design that still reaches print.

Study	Design	n	Duration	What was measured	Result
Human trials of collagen peptides with a bone outcome, as retrieved for this article
König 2018	RCT, double-blind, placebo	131	12 months	DXA T-score, P1NP, CTX-1	Spine T-score +0.1 vs −0.03 (p = 0.030)
Lampropoulou-Adamidou 2022	Randomised, no placebo; comparator was calcium and vitamin D alone	51	12 months	pQCT tibia, DXA spine and hip	Trabecular vBMD +2.54 ± 3.43% vs comparator (p = 0.02)
Zdzieblik 2021	Open-label follow-up, no control group	31	4 years	DXA	Increase reported; uncontrolled
Kuhlman 2026	Pilot RCT, placebo, female runners	22	4 weeks	Turnover markers only	P1NP interaction p = 0.04; CTX-1 not significant
Seyhan 2026	RCT, four arms including whey	40	8 weeks	DXA lumbar spine	+0.108 g/cm² in the whey-plus-collagen arm
Sun 2025	Meta-analysis of RCTs	—	—	BMD, turnover markers, muscle	BMD increased; I² = 80.1%

Not one of these trials counted a fracture

That table has no fracture column because no trial supplied one. A 2025 meta-analysis by Chao Sun's group at Lanzhou University pooled the randomised trials of collagen peptides, alone and combined with calcium and vitamin D, and found significant increases in femoral neck and spine density — with heterogeneity of I² = 80.1%. Heterogeneity that high means the trials are not converging on one effect; something differs between them that the pooled number papers over. Bone turnover markers moved by standardised mean differences of 0.40 to 0.58, and muscle performance by 0.60, with a confidence interval running from 0.05 to 1.15 — a lower bound close to nothing. The review lists fracture probability among its objectives and reports no pooled fracture estimate, for the straightforward reason that the underlying trials do not measure fractures.

This is the recurring trap in bone nutrition. Density is what assessment measures — DXA and REMS, or Radiofrequency Echographic Multi Spectrometry, both report density and the scores derived from it. A fracture is the thing density stands in for. Moving a proxy by a tenth of a standard deviation is a real finding about the proxy and an untested claim about anything else.

The quality spread is worth seeing plainly, too. An eight-week trial in 40 resistance-trained men reports a lumbar spine gain of 0.108 g/cm² in its whey-plus-collagen arm — roughly a tenth of a typical adult spine value, in eight weeks. Set that beside König's tenth of a T-score across twelve months in 131 women. Those two results cannot both be describing the same tissue. When a literature contains numbers that disagree by an order of magnitude, the small, short, single-site trials are the ones to discount.

The comparison that puts collagen in perspective is ordinary protein

Collagen is protein, and protein has its own bone literature, assembled to a higher standard. A systematic review and meta-analysis by Marissa Shams-White and colleagues, commissioned by the National Osteoporosis Foundation and covering 16 randomised trials and 20 prospective cohorts, found moderate-strength evidence that higher protein intake protects lumbar spine density — a net difference of 0.52% (95% CI 0.06 to 0.97, I² = 0%, from five trials). At the total hip, femoral neck and total body, no effect. Overall risk of bias, medium.

Taylor Wallace and Cara Frankenfeld, pooling cohort data across 29 studies, found a 16% lower rate of hip fracture comparing high with low protein intake. That is a relative figure, and the pooled analysis does not report the absolute fracture rates underneath it, so it cannot be translated into how many fractures that would mean for a given group of women. These are also cohorts, where people who eat more protein differ from people who eat less in many ways at once.

Against that background, 5 grams of collagen peptides is a small addition to a day's protein intake. And here is the structural problem running through every trial in the table above: with the partial exception of the whey arm in that eight-week study, they compare collagen against placebo, not against an equal dose of ordinary protein. A positive result therefore cannot separate collagen-the-signal from protein-the-nutrient. The trial that would settle the question has not been run.

What is settled, and what would settle the rest

Settled: collagen fragments reach human blood intact, measured directly by mass spectrometry. Settled well enough to act on, if skin is what you care about: a meta-analysis of 26 randomised trials in 1,721 participants found improvements in skin hydration and elasticity against placebo, though its authors flag several biases among the included trials.

Open: whether the bone signal is real at the size the trials suggest, and whether it changes anything a person would ever notice. What would close it is not complicated to describe — a blinded trial against an equal dose of non-collagen protein, run long enough for a full bone remodelling cycle, with fractures counted rather than inferred. Nobody has run it, and until someone does, the honest description of collagen and bone is a plausible signalling mechanism with twelve months of controlled human data on a proxy measure.

Which leaves a more useful question than which powder to buy. A supplement claiming to shift bone density by a tenth of a standard deviation is making a claim about a number — one that exists, that can be measured, and that can be measured again later. Knowing where that number actually sits, and which direction it has been moving, turns every supplement claim from something to believe into something to check.

This article is for education and is not a substitute for medical advice, diagnosis, or treatment. Talk to your physician about your individual situation.