Why animal data does not transfer to people

The claim an animal study is entitled to make

An animal experiment can establish that a compound does something in that animal, under those conditions, at that dose, by that route. That is a real finding and it is worth having.

It is not evidence that the compound does the same thing in a person. The step between the two is the entire drug development process, and most compounds do not survive it.

Almost every peptide sold outside the approved products rests on animal and cell-culture work. That is not a scandal, and it is not nothing. It is a specific tier of evidence with a specific ceiling, and the sections below are about where the ceiling comes from.

A dose in a rat is not a dose in a person

The most common misreading of an animal paper is arithmetic. A study reports a milligram-per-kilogram figure, someone multiplies by a body weight, and a number appears that looks like it means something.

FDA has a published procedure for this exact conversion, and it goes the other way. The agency starts from the highest dose in the animal study that produced no adverse effect. It converts that to a human equivalent by normalizing to body surface area rather than body weight, using a table of species factors. A mouse figure is divided by roughly twelve. A rat figure is divided by roughly six.

Then it picks a species. In the absence of data about which animal is the better model, the default is the most sensitive one, meaning the species that produces the lowest human equivalent.

Then it divides again. The default safety factor is ten, and the guidance says the factor should be raised when there is reason for increased concern and can be lowered only with data that provide added assurance.

What comes out of that is a maximum recommended starting dose for a first-in-human trial in monitored healthy volunteers. It is a ceiling on the first dose anyone is given under observation, arrived at by two successive reductions. It is not a dose for a reader, and no arithmetic on this page produces one.

FDA's own list of why species differ

The most useful passage in that guidance is the justification for the safety factor. It is a regulator writing down, in plain terms, what it expects to go wrong between an animal and a person.

Enhanced sensitivity to the compound's pharmacologic activity in humans compared with animals. Difficulties detecting certain toxicities in animals, and the examples the guidance gives are headache, muscle pain and mental disturbances. Differences in receptor density or receptor affinity. Unexpected toxicities. And differences in how the compound is absorbed, distributed, metabolized and eliminated.

Read that list against a peptide whose case rests on rodent studies. Three of the five items are about the target itself behaving differently, and one is about a whole category of effects an animal cannot report.

The guidance also notes that choosing the right species for a biological product involves factors unique to those products, such as whether the animal even expresses the relevant receptor. A molecule can be inert in a species that lacks its target and dramatic in one that has it.

What happened when someone checked

A systematic review published in the BMJ compared animal experiments with clinical trials for six interventions where the human answer was already unambiguous. It is the closest thing to a direct test of whether animal results predict human ones.

Two comparisons agreed. Thrombolysis improved outcomes in stroke patients, and in animal models tissue plasminogen activator reduced infarct volume by about a quarter. Bisphosphonates increased bone mineral density in patients, and alendronate increased it in animals.

Two disagreed, and the disagreements are the reason to read the paper. Corticosteroids showed no benefit in clinical trials for head injury, while animal models showed a clear benefit. Tirilazad was associated with a worse outcome in patients with ischemic stroke; in animal models it reduced infarct volume by 29 percent and improved neurobehavioral scores by 48 percent.

Those are not small effects that failed to replicate. They are large, confident animal results pointing at a treatment that turned out to be useless in one case and harmful in the other.

The authors' conclusion is worth carrying around. Discordance between animal and human studies may be due to bias, or to the failure of animal models to mimic clinical disease adequately.

The route breaks the number even between animals

Changing how a compound is delivered can invalidate an animal result without changing anything else, and FDA showed exactly that when it reviewed BPC-157 for compounding in July 2026.

Toxicity studies in rats and dogs established no-adverse-effect levels for the intramuscular route. The routes proposed for compounding were oral, rectal, subcutaneous and nasal.

The agency's finding was that no studies established the absolute bioavailability of the compound by the intramuscular route or the proposed ones. Because of that, it stated it was not possible to use the intramuscular no-adverse-effect levels to estimate the equivalents for the proposed routes.

That is a route failure inside the animal data, before the species gap is even reached. A safety number established one way into an animal did not carry to a different way into the same animal.

Duration is a design choice with a hard ceiling

A study that runs for a few weeks cannot detect what takes years, and the arithmetic of that translation is not intuitive.

FDA's Epitalon review sets it out. A standard rodent carcinogenicity study treats male and female animals daily for at least two years. The agency cites published work finding that a rodent study lasting twelve to eighteen months is equivalent to looking for human cancer in people aged thirty to fifty, and would therefore have markedly reduced sensitivity.

The studies actually available for Epitalon dosed mice for five consecutive days a month from two months of age, so total exposure reached at most about twenty-two weeks. The agency did not identify any two-year carcinogenicity study for either form of the substance.

Its conclusion named three specific limits rather than a general reservation. The studies used a fixed dose, assessed female mice only, and exposed the animals for a short period. On that basis the agency said the studies do not adequately inform either the genotoxic or the carcinogenic potential of the compound.

One dose level tells you nothing about the shape of a dose-response curve. One sex tells you nothing about the other. A few months tells you nothing about a lifetime.

Scope, and what a regulator says when it is not enough

The same reviews are useful for a second reason: they show what an agency actually says when it reads the animal literature for a peptide and finds it thin.

For BPC-157, repeat-dose toxicity studies ran twenty-eight days in rats and dogs. Longer studies were not available to show whether the findings reproduce or whether new signals appear with longer treatment. No studies were available to determine carcinogenic potential. No studies covered the proposed routes.

The agency also noted that dose-response relationships for the compound's reported effects had not been established, that its molecular targets had not been identified, and that its mechanisms remain poorly understood. It said this makes it difficult to assess the biological plausibility of the reported effects.

The wording it settled on was that nonclinical toxicological studies were too limited in scope and duration to inform safety considerations for the potential clinical uses proposed.

That is not a finding that the compound is dangerous, and it should not be read as one. It is a finding that the animal work available cannot answer the question being asked of it.

How to read an animal result on a product page

Check the species and the number of them. One species is a result; agreement across species is a stronger one, and FDA's own procedure assumes species will disagree.

Check the route. A finding established one way in does not carry to another way in, and this is the failure that recurs most often in the peptide literature.

Check whether more than one dose level was used. A single fixed dose cannot show a dose-response relationship, and without one an effect is hard to attribute.

Check the duration against the claim. A few weeks of exposure cannot support a claim about years of use, whatever the effect size was.

Check whether the mechanism is known. FDA leaned on this for BPC-157: without identified targets, an unexplained effect is harder to believe and harder to design around.

Then hold the result at the level it belongs. An animal finding is a reason to run a trial. It is not a substitute for one.

Sources

Each document below is named as it names itself, with the date printed on that document rather than the day it was read.

1. Estimating the Maximum Safe Starting Dose in Initial Clinical Trials for Therapeutics in Adult Healthy Volunteers — Guidance for Industry — U.S. Food and Drug Administration, July 2005
2. Comparison of treatment effects between animal experiments and clinical trials: systematic review — BMJ (Perel, Roberts, Sena and colleagues), January 2007
3. FDA Briefing Document for BPC-157-Related Bulk Drug Substances, Pharmacy Compounding Advisory Committee — U.S. Food and Drug Administration, May 2026
4. FDA Briefing Document for Epitalon-Related Bulk Drug Substances, Pharmacy Compounding Advisory Committee — U.S. Food and Drug Administration, May 2026