Peptide side effects: the counted and the uncounted

The short answer

It depends entirely on which peptide, and for most of them nobody knows. A handful of approved peptide medicines have adverse event rates measured in thousands of randomised patients. The rest, the ones sold as research chemicals, have no safety denominator at all, which is not the same as being safe.

A peptide is a short chain of amino acids, the same building blocks proteins are made of, just fewer of them linked together. Peptides carry signals between cells, which is why they can do useful things and also why they can do unintended ones: a signal rarely reaches only the tissue you meant it for.

An adverse event is the technical term for anything unwanted that happens to someone during a study, whether or not the drug caused it. A side effect implies causation; an adverse event is simply a thing that was written down, and working out which ones the drug caused is what a control group is for.

That is the fork this whole article turns on. For a small number of peptides, somebody ran that comparison at scale. A 2026 meta-analysis pooled four randomised trials of semaglutide involving 3613 participants and reported that treatment stopped because of adverse events in 6% of the semaglutide group against 2.9% on placebo [1]. For most peptides on the market, there is nothing of that shape. A 2025 systematic review of BPC-157 searched the literature back to database inception and ended with 36 studies, of which 35 were preclinical and one was clinical, and wrote plainly that no clinical safety data were found [5].

Both of those are answers to "what are the side effects". They are just very different answers, and the pages that list nausea and headache for both are answering neither.

What a side effect number actually needs

Three things: a count of how many people had the event, a count of how many people were in the group, and a comparison group that got nothing. Without the third, ordinary life gets scored as a drug effect, because people have headaches, nausea and tiredness whether or not they are injecting anything.

The clearest demonstration of why the comparison group matters sits inside the semaglutide meta-analysis itself. Add up its table of gastrointestinal adverse events across all four trials and 1614 of 2350 people on semaglutide reported one [1]. That is a striking number until you read across the same row, where 565 of 1263 people on placebo reported one too [1].

Nearly half of the group receiving a dummy injection reported the same category of symptom. Stomachs are not quiet places, and people paying close attention to their bodies for 68 weeks notice things [2]. The drug still comes out clearly worse once that background is subtracted, at a risk 1.62 times higher than placebo [1], and that ratio is the real finding. The raw percentage on its own would have wildly overstated it.

This is the single most useful thing to carry away from any safety claim about any peptide. A page that says "users report nausea, fatigue and headache" has described the placebo arm just as accurately as the treatment arm. It is not evidence of a side effect. It is a list of things human beings report.

The same meta-analysis is also honest about the ceiling of its own claim: the gastrointestinal events it pooled were, in the authors' words, generally short-lived, transient, and did not necessitate treatment discontinuation, while serious adverse events, including acute pancreatitis and cholelithiasis, were uncommon [1]. Cholelithiasis means gallstones. Uncommon is not the same as absent, and a rate is the only way to tell those apart.

What a counted peptide looks like

Semaglutide is the fully worked example. It went through large randomised trials with placebo arms, the adverse events were tabulated, and independent teams have since pooled them. That is why the numbers exist, and the numbers exist because somebody paid for 1961 people to be watched for 68 weeks.

The STEP 1 trial enrolled 1961 adults with obesity or overweight and no diabetes, and randomly assigned them in a 2:1 ratio to 68 weeks of once-weekly semaglutide at 2.4 mg or placebo [2]. Those three facts, how many, split how, for how long, are the machinery that produces a side effect rate. Everything downstream is arithmetic on them.

What that machinery produced, once pooled with three sibling trials, is a picture with edges. Gastrointestinal adverse events are common and clearly drug-related. Stopping treatment because of an adverse event happens to a minority, 6% against 2.9% on placebo, which is a real difference and also a reassuring one [1]. Serious events happen and are uncommon [1].

None of that makes semaglutide gentle. It makes it characterised. When a prescriber warns about nausea, that warning is standing on 3613 randomised participants rather than on a forum thread [1]. The difference is not that a characterised drug is safer. It is that somebody can tell you what you are accepting.

The same counting, a different peptide

Tirzepatide has been through the same process, and a 2023 meta-analysis of six randomised trials put per-symptom numbers on it. Reading them correctly means noticing the population: these were people with type 2 diabetes, and the comparison arms included other drugs rather than placebo alone.

A 2023 systematic review and meta-analysis pooled six randomised controlled trials with a total of 4,586 patients taking tirzepatide for type 2 diabetes [3]. Its comparators were a mix: placebo, other GLP-1 receptor agonist drugs, and insulin degludec. So these are not placebo-controlled rates and not weight-loss-population rates: they are diabetes-trial rates against whatever each trial used as its control.

With that caveat attached, the numbers are specific in a way no consumer page is. Nausea occurred in 20.43% of people on tirzepatide against 10.47% of comparators [3]. Diarrhoea in 16.24% against 8.63% [3]. Vomiting in 9.05% against 4.86% [3]. In every case the comparator rate is roughly half the drug rate, which is the shape you would expect from a real effect sitting on top of a real background.

The trial that established tirzepatide for obesity rather than diabetes was SURMOUNT-1, which assigned 2539 adults in a 1:1:1:1 ratio to 5 mg, 10 mg or 15 mg of tirzepatide or placebo for 72 weeks [4]. It is named here for the same reason STEP 1 was: not for its adverse event table, which its published abstract does not carry, but because the count and the split are what make a rate possible at all.

What BPC-157 has instead

BPC-157 is the most discussed research peptide in the world and has no clinical safety data. A 2025 systematic review found 36 studies, 35 of them in animals. The single clinical entry was a retrospective series of twelve knee patients, which is not a safety study by any definition.

The 2025 review searched PubMed, Cochrane and Embase from database inception to June 3, 2024, screened 544 articles from 1993 to 2024, and included 36 studies, of which 35 were preclinical and one was clinical [5]. The clinical one was a retrospective look at people who had received an injection into the knee joint for chronic pain, in which 7 of 12 patients reported relief for more than six months [5]. That is a report of benefit in twelve people with no control group, not a measurement of harm.

On safety the review is unambiguous. Preclinical work showed no adverse effects across several organ systems, and then the very next sentence reads: no clinical safety data were found [5]. The authors go on to say that adverse effects are possible due to unregulated manufacturing, contamination, or unknown clinical safety [5].

The nearest thing to a human safety study appeared the same year and is smaller than most people assume. It was a pilot at a private clinic in Florida with two participants, a 58-year-old man and a 68-year-old woman, each of whom had already received intravenous BPC-157 before the trial began [6]. They were given 10 mg infused over one hour on day 1 and 20 mg on day 2, and no effects on the tested markers and no reported side effects followed [6]. The authors conclude that the pilot showed the safety of BPC-157 in humans. Two people who were already users cannot show that, and the same paper says in its own background that few studies on humans have been published [6].

So the honest sentence about BPC-157 side effects is not "it has none". It is that nobody has counted, and an uncounted risk and an absent risk look identical from the outside.

The trial that was meant to count

One human safety trial of BPC-157 was registered in 2015 with adverse events as its primary outcome, and it has never reported. Three more registrations appeared in 2026, including a recruiting randomised placebo-controlled phase 2 in 120 people. None of the four has posted a result.

On the United States trial registry there is a phase 1 study of oral BPC-157 in healthy volunteers. Its design is exactly what the field is missing: tablets containing 1mg of BPC-157 or placebo, 42 participants planned, and a primary outcome recorded as, simply, adverse events [7].

It was first posted on 2015-12-22 and its record was last updated on the same date [7]. Results were expected to be submitted in May 2016 [7]. The registry today carries an overall status of UNKNOWN and no results [7]. A status of unknown means the sponsor has not confirmed to the registry that the study is still running or has finished.

Nothing about that proves anything went wrong. Small studies are abandoned for funding reasons constantly. What the record does show is a study whose primary outcome was adverse events, sitting untouched for a decade [7].

That changed this year, and it is the part almost no page about BPC-157 has caught up with. Three further registrations appeared on the same registry in 2026. The largest is a randomised, double-blind, placebo-controlled phase 2 trial of BPC 157 for accelerated repair of acute grade II hamstring strain, sponsored by Hudson Biotech, quadruple-masked, meaning participants, care providers, investigators and outcome assessors are all kept from knowing who received what, with 120 participants planned, first posted on 2026-02-27 and currently recruiting [12]. The second is a randomised, quadruple-masked, placebo-controlled pilot in 30 people after arthroscopic rotator cuff repair at the University of Arkansas, first posted on 2026-09-03 and not yet recruiting [13], whose own summary notes that the peptide has not yet been studied in formal human clinical trials [13]. The third is a completed single-group, unmasked study in 40 physically active adults taking BPC-157 gummies over an eight-week period, run by the company that sells them [14].

None of the four has posted a result [7][12][13][14]. So the honest position has two halves that have to be held together. The answer to "has anyone measured this in humans" is still no. The answer to "is anyone finally trying" became yes this year, and one of those attempts is exactly the randomised placebo-controlled design the field has been missing since 2015.

A case report is not a rate, and still matters

Outside trials, harm gets documented one person at a time. A case report can prove that something can happen without telling you how often. Treated as a rate it is useless; treated as an existence proof, it is sometimes the only evidence there is that a compound has a ceiling.

Melanotan II, a tanning peptide sold online, has no trial-grade safety dataset. What it has is case reports, and one from 2012 is a clean example of what they can and cannot tell you [8]. A 39 year-old man injected 6 mg of Melanotan II bought over the internet to darken his skin in winter; by his own account this was six times the recommended starting dose [8].

Two hours later he arrived at an emergency department with body aches, sweating and anxiety, a heart rate of 130 beats per minute that peaked at 146, dilated pupils and muscle tremors [8]. His creatine phosphokinase, a blood marker that rises when muscle tissue breaks down, was 1760 IU/L on arrival and 17773 IU/L twelve hours later, with kidney function impaired alongside it [8]. He spent three days in intensive care and recovered [8]. Crucially, the vial he injected was analysed by mass spectrometry and confirmed to be Melanotan II against a purchased reference standard [8], which rules out the usual objection that the problem was some other substance in the bottle.

That single report cannot tell you the odds. It has a numerator of one and no denominator whatsoever. What it can tell you is that this outcome is possible from this compound at a dose a person can buy and self-administer, and no amount of "no reported side effects" on a vendor page outranks a confirmed case. Our longer piece on Melanotan II and tanning nasal sprays covers what else is known.

The side effects that come from the vial, not the peptide

A large share of real-world harm from injected peptides is arithmetic. Vials, syringes and unit conversions produce overdoses that have nothing to do with the molecule, and regulators have documented people taking many times the dose they intended because of a measurement mistake.

In July 2024 the FDA issued an alert about compounded injectable semaglutide, and the mechanism it describes has nothing to do with pharmacology. It is about confusion between different units of measurement, and specifically between milligrams, millilitres and the marks printed on an insulin syringe, which are called units [9].

The reports the agency summarised describe patients drawing up five to 20 times more than the intended dose [9]. The version FDA illustrates is arithmetically simple and devastating: people were instructed to use a U-100 insulin syringe to draw a 5-unit dose, which is 0.05 millilitres, and instead administered 50 units [9]. Prescribers made the same class of error in the other direction, one intending 0.25 milligrams, which is 5 units, and prescribing 25 units, and another prescribing 20 units instead of 2 units, affecting three patients [9]. The resulting adverse events included nausea, vomiting, abdominal pain, fainting, headache, migraine, dehydration, acute pancreatitis and gallstones, some requiring hospitalisation [9].

Every one of those is recorded as a side effect and none of them is a property of semaglutide. They are properties of a multi-dose vial, a syringe scaled for insulin, and a conversion nobody checked. That risk applies with far more force to research peptides, which arrive as a powder that the buyer reconstitutes themselves at whatever concentration they chose. If you are going to be anywhere near that arithmetic, our reconstitution guide and calculator exists to make the conversion explicit, and the guide to vetting research peptides covers what is in the vial before you get to how much of it you drew.

What shows up once a peptide is everywhere

Poison centre data catches what trials cannot: what happens when a drug leaves the clinic and reaches millions of ordinary households. For GLP-1 drugs the answer is that most calls are dosing mistakes, most outcomes are mild, and the volume rose sharply as prescribing did.

A 2024 analysis of the United States National Poison Data System looked at every single-substance GLP-1 case reported to poison centres from 2017 to 2022 and found 5,713 of them [10]. The profile is not what a drug-scare framing would predict. Most cases were among females, at 71.3%, and attributable to therapeutic errors, at 79.9% [10]. A therapeutic error means someone took their own medicine wrongly: the most frequent scenario recorded was inadvertently taking or being given a dose twice.

Outcomes were mostly unremarkable. 22.4% of cases were evaluated in a healthcare facility, including 0.9% admitted to critical care and 4.1% admitted elsewhere in hospital, and serious medical outcomes were described in 6.2% of cases, including one fatality [10]. One death in nearly six thousand reports over six years is a real number that deserves neither dismissal nor headline treatment.

The trend line is the part worth keeping. The rate rose from 1.16 cases per million people in 2017 to 3.49 in 2021, then jumped 80.9% to 6.32 in 2022 [10]. That climb is in the paper's own abstract, framed as a rapid increase rather than as a new hazard [10]. This is also why no comparable dataset exists for research peptides: nobody is systematically collecting one.

Immunogenicity, the risk a small molecule does not carry

Peptides are made of the same material as the proteins your immune system polices, so your body can learn to recognise one and attack it. What raises that risk is often not the peptide itself but the fragments and by-products left over from manufacturing.

Immunogenicity means a drug provoking an immune response against itself. The body produces antibodies to the medicine, which can blunt its effect, change how quickly it is cleared, or in rarer cases cause a reaction in its own right. Aspirin cannot do this. A peptide can, because it is built from amino acids and that is precisely the material immune surveillance is tuned to.

The interesting part, for anyone buying peptides outside a pharmacy, is where the risk concentrates. A 2023 review of immunogenicity risk assessment for synthetic peptide drugs notes that of the 80 peptide drugs approved for use in humans, at least five are now off-patent and are being made as generics, and that regulators ask generic sponsors to characterise new impurities specifically because the impurities can have potential to elicit unwanted immune responses owing to the introduction of T-cell epitopes [11]. An epitope is the small fragment an immune cell actually recognises.

Read that again with a research-peptide vial in mind. The concern serious enough to shape regulatory pathways for approved generics is about the leftovers of synthesis, not the intended molecule [11]. A product with no certificate of analysis has no characterised impurity profile, which means the one variable this entire risk assessment framework is built around is simply unknown. That is a different and larger worry than whether the peptide on the label has side effects.

How to read any peptide side effect claim

Ask three questions of any safety statement: out of how many people, compared with what, and who was counting. A claim that cannot answer all three is not a safety profile. It is a list, and lists of symptoms are cheap because everybody has symptoms.

Put the two halves of this article side by side. For semaglutide and tirzepatide you can name the trials, the enrolment, the comparison arms and the pooled rates, and disagree about what they mean. For BPC-157 you can name a decade-old registration that went silent, three newer ones that have not reported anything yet, and a pilot in two people. Both of those are the current state of the evidence. Only one of them supports any sentence beginning "the side effects are".

So when a page tells you a peptide causes injection site reactions and mild nausea, the useful response is not to believe it or disbelieve it but to ask where the number came from. If there is no number, there is no finding. If there is a number with no comparison group, roughly half of it probably belongs to being a person. And if the peptide has never been through a trial at all, the absence of reported side effects is telling you about the state of the literature, not about the state of your body.

That is also why what is in the vial and how much of it you drew matter so much. For the peptides nobody has characterised, manufacturing quality and measurement arithmetic are the two risks you can actually do something about.