Reading a peptide study: the evidence ladder
Last updated: October 27, 2026 · 10 min read · By the Grey Peptides Editorial Board
- Research comes in rungs. A cell study shows something can happen; an animal study shows it can happen in a body; only controlled human trials show whether it happens in people, and how often.
- Many research peptides stop on the lower rungs. BPC-157 has 228 PubMed records and 0 indexed as randomised trials; semaglutide has 313.
- Four questions sort almost any paper: who or what was studied, was there a comparison group, how many and for how long, and what exactly was measured.
Why 'backed by research' means so little
A peptide can be described as researched whenever papers about it exist. What the phrase hides is what kind of papers. A study of cells in a dish, a study of rats, a report on two patients and a randomised trial in two thousand people are all 'research'. They answer different questions, and only the controlled trials in people answer the question a buyer is asking: will this do what is claimed, in people like me, and at what risk?
Evidence-based medicine has long arranged evidence as a pyramid, with the weakest kinds at the bottom and systematic reviews and meta-analyses at the top, though some argue reviews are better seen as a lens for reading the other rungs than as a rung of their own 1. The rungs below follow that order, each with a real paper about a peptide, chosen from studies already cited on our encyclopedia entries. The point is not that the lower rungs are worthless. The point is that a claim can only be as strong as the highest rung it reaches.
Rung 1: cells in a dish
In vitro studies, Latin for 'in glass', put a peptide on cells grown in a dish and measure what changes. They are good at showing mechanism: which receptor a molecule binds, which genes it switches on. A 2014 study found that BPC-157 increased growth hormone receptor expression in fibroblasts taken from rat Achilles tendon, in a dose- and time-dependent way 2, and an earlier one found that it sped the outgrowth and migration of tendon cells from rat Achilles tendon explants 3.
What a cell study cannot tell you is whether any of it happens in a living body. A dish has no digestion, no blood flow, no liver breaking the molecule down and no immune system reacting to it, and the concentration is whatever the researcher chooses, not what an injection delivers to the tissue. Claims that a peptide 'switches on genes' or 'upregulates growth factors' usually describe this rung; the 2014 paper itself concluded only that the effect 'may' contribute to tendon healing 2.
Rung 2: animals
Animal studies give the peptide to a whole organism, usually a mouse or a rat, often after deliberately injuring it. They add the things a dish lacks: absorption, distribution, metabolism and a body's response. In one, rats with crushed calf muscles healed better on functional, biochemical and microscopic measures when given BPC-157, a study run by the Zagreb group that first described the peptide 4 5. Studies like this are a real step up, and they are the backbone of the BPC-157 literature: 106 of its 228 PubMed records are indexed as animal-only research 6.
The limits are worth stating. Rodents differ from people in metabolism, body size and how they heal, and injuries made in a laboratory are uniform in a way human injuries are not. An animal result is a reason to test in people, not evidence that people will benefit.
Rung 3: case reports and series
The first human evidence is often a description of what happened to a handful of patients, with no comparison group. In a 2025 pilot at a private clinic in Florida, two adults who had received intravenous BPC-157 before were given 10 mg and then 20 mg, and the authors reported no changes in heart, liver, kidney, thyroid or glucose markers 7. That is a first look at safety, in two people.
What it cannot show is effectiveness, or rare harms. With two people, a side effect that strikes one person in a hundred would be missed about 98 times in 100 (0.99 squared, computed here). Without a comparison group, any improvement could be due to time, rest, other treatment or expectation. A 2026 report of three dental patients given TB-500 with BPC-157 and other drugs makes the same point in its own words: with no controls, spontaneous recovery cannot be excluded and causality cannot be established 8. Testimonials and before-and-after photos sit on this rung or below it.
Rung 4: small controlled trials
A controlled trial compares people who get the treatment with people who do not, ideally assigned by chance, which is what 'randomised' means, and ideally without either side knowing who got what, which is 'blinding'. Even a small randomised trial is a step up, because chance assignment balances the things researchers cannot measure.
GHK-Cu, the copper peptide in many skin products, has a small randomised trial that illustrates both the strength and the limit. Thirteen patients completed a randomised trial of skin care after laser resurfacing with or without GHK-Cu. Computer analysis and blinded evaluators found no difference in how quickly redness resolved, and no difference between groups in wrinkles or overall skin quality at 12 weeks, although patients using GHK-Cu rated their own skin higher on a questionnaire 9. That is a real test, and the split between the blinded measures and the patients' own ratings is exactly the kind of detail an abstract can hide. But with thirteen people, a modest benefit could also have been missed. Early safety trials sit here too: thymosin beta-4, the full protein from which the TB-500 fragment is taken, was given intravenously to 40 healthy volunteers in a randomised, placebo-controlled phase 1 study designed to test safety and tolerance, not benefit 10. FDA's May 2026 review of BPC-157 lists a randomised, double-blind, placebo-controlled study in ulcerative colitis, but it was reported only as a 2005 meeting abstract and FDA judged its data inadequate to support efficacy or safety 11: a trial that is not fully published cannot be fully read.
Rung 5: large randomised trials
At the top of primary research are large randomised controlled trials, usually phase 3, designed to answer a specific question with enough people to be sure. STEP 1 randomised 1,961 adults without diabetes to weekly semaglutide or placebo; after 68 weeks, average weight loss was 14.9% against 2.4% 12. SELECT went further, randomising 17,604 adults aged 45 or older with heart disease and overweight but no diabetes, to test whether semaglutide reduced heart attacks, strokes and cardiovascular deaths; it cut those events by 20% over a mean follow-up of 39.8 months 13. Both were funded by Novo Nordisk, the drug's maker, and both list Novo Nordisk employees among their authors 12 13.
The gap between approved and research peptides is wide on this rung: semaglutide has 313 PubMed records indexed as randomised trials; BPC-157 has 0, TB-500 has 0 (the human trials quoted for it tested the full thymosin beta-4 protein, not TB-500), and GHK-Cu has 3 6.
Rung 6: systematic reviews and meta-analyses
Above single trials sit systematic reviews, which search for every study on a question using a stated method, and meta-analyses, which pool the trials' numbers 1. They show whether findings hold up across many studies. A 2021 meta-analysis of 19 randomised, double-blind, controlled trials in 1,125 people, 95% of them women, found that hydrolysed collagen improved skin hydration, elasticity and wrinkles compared with placebo 14.
But a review is only as good as what it pools, and the word 'systematic' does not lift weak studies. A 2025 systematic review of BPC-157 in orthopaedic sports medicine searched three databases and included 36 studies: 35 preclinical and one clinical, a retrospective series in which 7 of 12 patients given knee injections reported relief for more than six months; it found no clinical safety data, and noted that clinicians and athletes use the compound anyway 15. A 2025 review of GHK as an anti-wrinkle peptide found a surprising absence of clinical studies of GHK-Cu and its palmitoylated form 16. Read the conclusion of a review for what it says about the evidence, not just for its title.
Four questions that sort any paper
Who or what was studied? Cells, animals or people, and if people, which people. A result in young athletes may not apply to older adults; a result in mice does not apply to anyone yet. The abstract usually says in its first lines.
Was there a comparison group, and was it randomised and blinded? Without a comparison, improvement cannot be credited to the treatment. Without randomisation, the groups may differ in ways that explain the result. Without blinding, expectation can.
How many, and for how long? Small numbers miss real effects and rare harms; short follow-up misses what happens later. Two patients for a few days is a different thing from two thousand for a year.
What exactly was measured? A blood marker, a scan or a questionnaire is not the same as a person feeling better, healing faster or living longer. Trials often report many outcomes; look for the one named in advance as primary, and whether it was met.
Two extra checks are worth the time. Who paid for the study, and who ran it? Industry funding and discoverer-led research are not disqualifying, but they are reasons to look for independent replication. And is the paper still standing? Our encyclopedia checks every cited paper for retractions 17.
Red flags in how research is cited
When a study is cited to support a claim, the citation itself often gives the game away. Watch for a claim about people backed only by a paper about rats or cells; for a 'study' that turns out to be a review by the compound's own discoverers, summarising their earlier work; for a link to a conference abstract that was never published in full; and for numbers that appear nowhere in the paper cited. Our dihexa entry shows another trap: the 2014 paper that tied dihexa to its proposed mechanism is listed by PubMed as retracted. And FOXO4-DRI shows how much attention a single mouse paper can draw with no human trial behind it.
The fix is simple, if tedious: open the paper, or at least its PubMed abstract, and check that it says what it is said to say, about the species it is said to be about. Every study card on our encyclopedia links to its PubMed record for exactly that reason.
How this maps to our grades
Our evidence grades compress the ladder into three levels, and they grade the quality of the human evidence, not whether a compound works 17. A high grade means efficacy has been tested in controlled human trials large enough to answer the question, typically a completed phase 3 programme or an approval dossier; 178 of our 696 entries are graded high. Medium means people have taken it under study conditions with published results, but short of a completed phase 3; 216 entries. Low means there is no human efficacy evidence worth relying on, only cell, animal or case-report evidence; 302 entries, including BPC-157, TB-500 and GHK-Cu 17.
A high grade is not a recommendation, and several high-graded compounds failed their main trials: the grade describes how well the question was tested, not the answer 17. Our Evidence Explorer shows the shape of the literature behind every entry, counted rung by rung, and our methodology page sets out how grades are assigned.
Frequently asked questions
Is BPC-157 proven?
No. Its research is mostly in cells and animals; of 228 PubMed records, 0 are indexed as randomised controlled trials. FDA's May 2026 briefing lists five clinical studies, including a randomised colitis study known only from a meeting abstract, and judged the data inadequate. We grade its human evidence low.
What is the strongest kind of evidence?
Large randomised controlled trials, and systematic reviews or meta-analyses that pool several good trials. Cell and animal studies come first in development but cannot show that a treatment helps people.
Does an animal study mean a peptide works in humans?
No. Animals differ from people in metabolism, size and healing, and laboratory injuries are uniform. An animal result is a reason to test in people, not evidence that people will benefit.
What does randomised and blinded mean?
Randomised means chance decides who gets the treatment, which balances the groups; blinded means participants, and ideally researchers, do not know who got what, which stops expectation colouring the results.
Related on Grey Peptides
Sources
- Murad, M. H., et al. (2016). New evidence pyramid. Evid Based Med, 21(4), 125-7. PMID: 27339128
- Chang, C. H., et al. (2014). Pentadecapeptide BPC 157 enhances the growth hormone receptor expression in tendon fibroblasts. Molecules, 19(11), 19066-77. PMID: 25415472
- Chang, C. H., et al. (2011). The promoting effect of pentadecapeptide BPC 157 on tendon healing involves tendon outgrowth, cell survival, and cell migration. J Appl Physiol (1985), 110(3), 774-80. PMID: 21030672
- Novinscak, T., et al. (2008). Gastric pentadecapeptide BPC 157 as an effective therapy for muscle crush injury in the rat. Surg Today, 38(8), 716-25. PMID: 18668315
- Sikiric, P., et al. (2024). New studies with stable gastric pentadecapeptide protecting gastrointestinal tract. Significance of counteraction of vascular and multiorgan failure of occlusion/occlusion-like syndrome in cytoprotection/organoprotection. Inflammopharmacology, 32(5), 3119-3161. PMID: 38980576
- Grey Peptides evidence counts (_ops/evidence_counts.json): PubMed records by study type and species for each entry, with queries, re-read October 27, 2026.
- Lee, E., et al. (2025). Safety of Intravenous Infusion of BPC157 in Humans: A Pilot Study. Altern Ther Health Med, 31(5), 20-24. PMID: 40131143
- Froum, S., et al. (2026). Case Report: Regenerative peptides for Inferior alveolar nerve injury: a narrative review with hypothesis-generating three clinical cases. Front Oral Health, 7, 1907109. PMID: 42876687
- Miller, T. R., et al. (2006). Effects of topical copper tripeptide complex on CO2 laser-resurfaced skin. Arch Facial Plast Surg, 8(4), 252-9. PMID: 16847171
- Ruff, D., et al. (2010). A randomized, placebo-controlled, single and multiple dose study of intravenous thymosin beta4 in healthy volunteers. Ann N Y Acad Sci, 1194, 223-9. PMID: 20536472
- U.S. Food and Drug Administration. FDA Evaluation of BPC-157-Related Bulk Drug Substances (BPC-157 (free base) and BPC-157 acetate), briefing document for the July 23-24, 2026 Pharmacy Compounding Advisory Committee, dated May 11, 2026: five clinical studies, including a randomised ulcerative colitis study reported as a 2005 meeting abstract. Read October 28, 2026. FDA
- Wilding, J. P. H., et al. (2021). Once-Weekly Semaglutide in Adults with Overweight or Obesity. N Engl J Med, 384(11), 989-1002. PMID: 33567185
- Lincoff, A. M., et al. (2023). Semaglutide and Cardiovascular Outcomes in Obesity without Diabetes. N Engl J Med, 389(24), 2221-2232. PMID: 37952131
- de Miranda, R. B., et al. (2021). Effects of hydrolyzed collagen supplementation on skin aging: a systematic review and meta-analysis. Int J Dermatol, 60(12), 1449-1461. PMID: 33742704
- Vasireddi, N., et al. (2025). Emerging Use of BPC-157 in Orthopaedic Sports Medicine: A Systematic Review. HSS J, 21(4), 485-495. PMID: 40756949
- Mortazavi, S. M., et al. (2025). Topically applied GHK as an anti-wrinkle peptide: Advantages, problems and prospective. Bioimpacts, 15, 30071. PMID: 39963574
- Grey Peptides methodology: how evidence is graded (including that several high-graded compounds failed their endpoints), grade counts from the encyclopedia dataset, and the retraction check on every cited PMID. Read October 28, 2026. Grey Peptides
Educational information, not medical advice. Each dose in this guide names its source, an approved label or a published study. None is a recommendation for you. An unapproved compound has no established safe or effective human dose, and products sold for “research use only” are not made or tested for people. Talk to a doctor before acting on anything on this site, including before you start, stop or change any medicine or dose.
New guides and rule changes, by email
One short email when a peptide's legal status moves — an FDA or PCAC decision, a WADA list change, an enforcement action — plus new guides and tools. At most one a week, and none when nothing changes. The newsletter has not started sending yet: your first email will be its first issue.
We keep your email address and the page you signed up from, nothing else. How it works · Privacy · RSS instead