Cognitive Peptides, Fact-Checked: Semax, Selank, Noopept, Dihexa, Cerebrolysin and P21

The cognitive peptide market sells six things that are not the same kind of thing. One is a registered Russian medicine with a real but untranslated clinical literature. One is a porcine brain extract with 41 English-language randomised trials behind it and a Cochrane review that is not kind to them. One is built on papers a journal has retracted for fabricated data. One has never been given to a human being in any published study. They are sold on the same shelves, at similar prices, with similar copy.

This page separates them by what can be checked. Every figure links to the document it came from, and every document was opened on 2026-09-05. Where the record is silent, or a source blocked us, the page says so rather than filling the gap. Nothing here is a test result of ours, no product on this page has been through our testing protocol, and there is no ranking here — vendors' published claims are graded against the record, not the vendors against each other. How brands appear on this site is set out on our disclosure page.

The six, at a glance

The last column is the point of the page: for four of the six, the most important document is one that records something missing.

Semax: FDA read the file and could not find the pharmacology

Semax has the most complete public paper trail of any compound here, because FDA had to build one. Its reviewers went through everything the nominator submitted and everything else they could find, and published the result as a briefing document for the Pharmacy Compounding Advisory Committee meeting of July 23-24, 2026. Four findings from it are worth quoting exactly.

No human pharmacokinetics, by any route. "We were not able to find pharmacokinetic studies in humans following exposure to semax (free base) or semax acetate via any ROA." Semax has been marketed for decades, and FDA's reviewers could not find a study of what the human body does with it in the literature available to them.

No safety data for one of the two nominated routes. Both were nominated: "The semax-related drug products proposed in the nominations are intranasal spray or subcutaneous injection ..." FDA could assess only one of them: "We were unable to find literature that discussed administration via subcutaneous injection ROA. The only ROA discussed in the literature was intranasal. Therefore, assessment of safety for semax was limited to the intranasal ROA."

"Nootropic" is not a medical condition. Semax was nominated for several uses including "nootropic". FDA declined to evaluate it as a distinct indication: nootropic "does not have an ICD-10 code and no professional society treatment guidelines could be found for this use", so it was folded into the cerebral ischaemia review. There is no regulatory object called cognitive enhancement in healthy people for a substance to be effective at.

The effectiveness evidence is two papers, and neither is good. For cerebral ischaemia, the only assessable reference was a conference abstract with an unknown number of human subjects and no reported clinical outcomes. For migraine and trigeminal neuralgia, one 1996 study in which, per FDA's summary of the authors' own conclusion, "no significant changes in TSEP were observed" and semax "does not exhibit analgesic activity by itself". FDA's overall conclusion: "Only two available references were available with insufficient details on the design and conduct of the studies, and the available references demonstrated lack of effectiveness."

FDA's proposal was explicit: "Accordingly, we propose not adding semax (free base) or semax acetate to the 503A Bulks List." On 24 July 2026 the committee voted 8-5 the other way, as RAPS reported that day.

Two further details belong on the record. FDA's search of the FAERS adverse-event database through 3 December 2025 returned exactly one report for semax: a consumer who reported ocular pain and burning after using semax nasal drops bought online, with hospitalisation, the pain unresolved a year later. And semax "is a registered drug in Russia and is available as 0.1% and 1% nasal drops" — true, and not the same as having been shown to do anything for cognition in healthy people. A large Russian-language clinical literature exists; FDA excluded it under 21 CFR 10.20(c)(2) for want of verified English translations. That is a procedural exclusion, not a finding that the studies are bad — but it does mean no reader of English can check them, which is the situation a buyer is in.

Selank: the same shape of evidence, without the review

Selank is a synthetic heptapeptide analogue of tuftsin, registered in Russia as an anxiolytic. Its clinical base is a handful of small Russian trials: a 62-patient comparison against medazepam in generalised anxiety disorder and neurasthenia (Zozulia et al., 2008), a 60-patient comparison against phenazepam in anxiety and somatoform disorders (2014). Both are indexed in MEDLINE; both are in Russian. On a Europe PMC title search on 2026-09-05, MEDLINE holds 59 selank records. Two are tagged randomised controlled trials. Neither is in English. None is a trial of cognition in healthy adults.

Selank was not among the seven substances the advisory committee reviewed in July 2026. It went through the earlier stage and stopped: FDA placed selank acetate in Category 2, the list of nominated bulk substances that "may present significant safety risks", writing that compounded drugs containing it "may pose risk for immunogenicity for certain routes of administration due to the potential for aggregation and peptide-related impurities" and that "FDA lacks important information regarding any safety issues raised by selank acetate administered to humans." It then appears in the sub-list of substances "previously in category 2 of the interim policies [that] were withdrawn by the nominators." The nomination was pulled, which is why FDA's November 2021 warning letter to a compounding physician could say that selank and semax "were not nominated for inclusion on the 503A bulks list": by then nothing was pending. The compound is still sold.

Noopept: a dipeptide in the nootropic aisle, not the peptide aisle

Noopept (omberacetam, N-phenylacetyl-L-prolylglycine ethyl ester) is the odd one out: a dipeptide derivative sold as a powder or capsule alongside racetams rather than injectables, and registered as a medicine in Russia for cognitive disorders of vascular and traumatic origin. MEDLINE holds 61 title records: 36 English, 25 Russian. The English ones are overwhelmingly rodent pharmacology. The human clinical work — neurasthenia, post-traumatic mild cognitive impairment, a comparison against piracetam — is in Zhurnal nevrologii i psikhiatrii imeni S.S. Korsakova, in Russian. None of the 61 is tagged a randomised controlled trial. One, a 2021 Heliyon paper on noopept and spinal microglia, was retracted the same year.

Noopept also supplies the cleanest example here of the gap between selling copy and record. FDA's February 2022 warning letter to Crystal Clear Supplements quotes the seller's own product page back at it as evidence the product was an unapproved new drug: "Noopept powder is a nootropic with a molecular structure similar to racetams, known to be 1000 times more potent than the parent racetam, piracetam. Noopept is fast acting and provides a cognitive boost. . . has mild anxiolytic effects. . . . neuroprotective . . . treat Alzheimer's disease, meaning it has positive effects on memory and learning abilities." That is the seller's sentence, not FDA's, and FDA's use of it is the point: copy like that is what converts a chemical into a drug in the agency's reading. The same pattern runs through every letter in our FDA warning-letter tracker.

Dihexa: the mechanism paper is retracted, and so is most of what is left

Dihexa is where the market's evidence problem stops being about thin data and starts being about data found to be false.

In October 2012, Washington State University issued a press release about work from Joseph Harding's laboratory reporting dihexa to be "seven orders of magnitude more powerful than BDNF" at forming new synapses — "It would take 10 million times as much BDNF to get as much new synapse formation as Dihexa." That is the origin of the "10 million times more potent than BDNF" line still circulating in the market today. Between 2011 and 2014 the laboratory published four papers in the Journal of Pharmacology and Experimental Therapeutics establishing dihexa's identity and mechanism. In September 2021 all four received Notices of Concern about possible image manipulation. In April 2025, three were retracted. The notice for the 2014 mechanism paper — the one establishing that dihexa works through the HGF/c-Met system at all — reads in full: "This article has been retracted at the request of the Editor. Following an investigation by Washington State University, Figures 1B, 2A/C, and data in the subsequent erratum submission for the article have been found to contain falsified and/or fabricated data and Leen H. Kawas and Joseph W. Harding were found to be solely responsible." The 2011 and 2012 papers went the same month. The 2013 paper, which introduced dihexa itself and reported the potency comparison, still carries only the 2021 Notice of Concern in MEDLINE as of 2026-09-05.

In January 2025 the Department of Justice announced that Athira Pharma agreed to pay $4,068,698 to resolve False Claims Act allegations that, between 2016 and 2021, it failed to report to NIH the allegations that its then-CEO had falsified and manipulated images in her doctoral dissertation and in papers referenced in grant applications. Secondary reporting also states that Washington State University revoked the doctorate; we could not open a WSU statement confirming it, so we report it as reported, not as verified.

The clinical successor was built and tested. Athira's fosgonimeton (ATH-1017), from the same programme, entered a Phase 2/3 trial in mild-to-moderate Alzheimer's disease. On 3 September 2024 the company reported that in 312 patients the primary endpoint — a global statistical test combining ADAS-Cog11 and ADCS-ADL23 at 26 weeks — was not met (p=0.70), and neither key secondary endpoint reached significance. That is the closest thing to a clinical test the dihexa hypothesis has had, and it failed.

What remains for dihexa itself is two MEDLINE title records, both 2021, neither a clinical trial — plus FDA's summary in the Category 2 list: "FDA has not identified any human exposure data on drug products containing dihexa acetate administered via any route of administration." The Alzheimer's Drug Discovery Foundation's Cognitive Vitality review, updated 13 August 2021, is blunter on the specific hazard — "Clinical trials: none to date. Observational studies: none to date," and "Theoretically, dihexa via activation of HGF and c-Met could promote tumorigenesis and cancer progression" — and records the finding least likely to appear on a nootropic product page: "Dihexa did not improve cognitive functions in rats with normal cognition." Even in the animal work, the effect was in impaired animals.

Cerebrolysin: the one with real trials, and the one Cochrane looked at

Cerebrolysin is not a peptide. It is a porcine brain enzymatic hydrolysate — a mixture of low-molecular-weight peptides and amino acids whose composition is defined by a manufacturing process rather than a sequence. It cannot be characterised the way a synthetic heptapeptide can. It was not among the seven substances at the July 2026 committee, and the reason on the record is simpler than a framework question: FDA's November 2021 warning letter to a compounding physician states that cerebrolysin "w[as] not nominated for inclusion on the 503A bulks list", and cerebrolysin does not appear anywhere on FDA's Category 2 page as we read it on 2026-09-05. It does have, by a wide margin, the most human data of anything here: 440 MEDLINE title records, 41 English-language randomised controlled trials, 42 studies registered on ClinicalTrials.gov. It is the one compound on this page where "does it work?" can be asked of pooled evidence.

The largest single trial is CASTA (Stroke, 2012): 1,070 patients with acute ischaemic hemispheric stroke randomised within 12 hours to cerebrolysin or saline placebo on top of aspirin, followed 90 days, primary endpoint a combined global test of modified Rankin Scale, Barthel Index and NIHSS. In the authors' words: "The confirmatory end point showed no significant difference between the treatment groups." A favourable trend in a post-hoc severity subgroup is reported, with a call for confirmation.

Cochrane's 2023 review (Ziganshina et al., CD007026.pub7, 11 October 2023) included seven trials and 1,773 participants, though every individual outcome rests on fewer than that. On mortality: "Cerebrolysin or Cortexin probably result in little to no difference in all-cause death" (RR 0.96, 95% CI 0.65 to 1.41; 6 trials, 1,689 participants; moderate certainty) — the subject is both agents because one pooled trial tested Cortexin, a related preparation, rather than cerebrolysin. On safety, from 3 trials and 1,335 participants: probably little to no difference in the total number of people with serious adverse events (RR 1.16, 95% CI 0.81 to 1.66; moderate certainty), but within that "an increase in the total number of people with non-fatal SAEs" (RR 2.39, 95% CI 1.10 to 5.23; moderate certainty), more prominent in the higher cumulative-dose schedule. The review also records that "the manufacturer of Cerebrolysin supported three multicentre studies, either totally, or by providing Cerebrolysin and placebo, randomisation codes, research grants, or statisticians."

So: a large, manufacturer-linked trial programme in stroke, a null primary result in the biggest trial, no mortality benefit on pooling, a signal of harm in non-fatal serious adverse events. None of it is about cognitive enhancement in healthy people, which is what the nootropic market sells it for.

P21: sold, never given to a human

P21 (P021) is a tetrapeptide mimetic of an active region of ciliary neurotrophic factor, developed in an academic laboratory and studied in mouse models of Alzheimer's disease and tauopathy. The preclinical work is real and continues, but it is small: a Europe PMC title search of MEDLINE on 2026-09-05 returns seven records for "P021", and only two of them are about the compound — a diffusion-MRI paper on early P021 treatment in a mouse model, and its corrigendum. The other five are conference-abstract numbers on unrelated topics.

There is no human study. Not a Phase 1, not a pharmacokinetic study, not a case series. A ClinicalTrials.gov search on 2026-09-05 returns five records matching "P021", none involving the compound — they are fuzzy matches on identifiers in unrelated trials. We looked for P21 on FDA's 503A and 503B nomination lists and did not find it, which means it has not even reached the stage where FDA writes down what is missing. It is nevertheless sold by multiple vendors with cognitive framing. A buyer of P21 is buying a molecule whose entire human record is empty.

What the July 2026 vote did, and did not, do

Section 503A of the Federal Food, Drug, and Cosmetic Act (21 U.S.C. 353a) lets a pharmacy compound a drug from a bulk substance only if that substance complies with an applicable USP or NF monograph; or, if no monograph exists, is a component of an FDA-approved drug; or, if neither, "appear[s] on a list developed by the Secretary through regulations". That third route is the 503A bulks list, and it is what the committee was voting about.

On 23-24 July 2026 the Pharmacy Compounding Advisory Committee considered seven nominated peptide substances: BPC-157, KPV, TB-500, MOTS-c, emideltide (DSIP), semax and epitalon. It recommended six and voted against emideltide. Semax passed 8-5, epitalon 7-5 with one abstention, emideltide failed 6-7 with one abstention.

What that is not: an approval, an addition to the list, or authorisation to compound anything. As Mintz's analysis put it on 29 July 2026, "an advisory committee vote is not an agency action, and FDA officials will have to make an ultimate decision on whether to accept or reject those recommendations," and "advisory committee recommendations are just that – advisory – and as such are not binding on FDA." Adding a substance requires notice-and-comment rulemaking.

Three things follow for a buyer. The vote covers exactly one of the six compounds here: selank and dihexa were nominated and withdrawn; noopept, cerebrolysin and P21 were never in this process. The vote went against FDA's own scientific review, which is not evidence the science is better than FDA said — the briefing document is still the record. And compounding is already happening ahead of any rulemaking: on 2026-09-05 we opened a retail page offering a compounded semax/selank nasal spray at $150 per spray plus a $39 telehealth visit, describing it as compounded by "Optimal Balance Pharmacy, a 503A pharmacy in Texas" and "dispensed for human use", while also stating that "neither peptide is FDA-approved in the United States". We report the listing as we read it; whether that arrangement satisfies section 503A is a legal question this page does not answer.

What vendors claim versus what is documented

Every claim below was read on a live page on 2026-09-05 and is quoted as written.

Medibact's bacteriostatic water is one of the products listed in our bacteriostatic water guide, where it is ranked third of four against that page's published criteria.

The pattern is simple. Purity claims are about the powder. Potency claims are usually about a cell assay. Neither is a claim that the compound changes cognition in a person, and on this shelf almost nobody makes that claim in a form that could be checked, because the studies that would support it do not exist.

You cannot verify any of this on yourself

This is the section that matters most to the person reading this page, because the natural response to an evidence vacuum is to run the experiment personally: take a cognitive test, take the compound, take the test again. That does not work, and the reason is arithmetic rather than opinion.

Taking a test twice raises the score. Calamia, Markon and Tranel's meta-analysis in The Clinical Neuropsychologist (2012) pooled nearly 1,600 effect sizes and found scores rise on retest across essentially every neuropsychological domain. Hausknecht and colleagues, pooling 107 samples and 134,436 participants in the Journal of Applied Psychology (2007), reported an adjusted overall effect size of 0.26, larger with identical forms. In the case that matters most, Estevis, Basso and Combs gave the WAIS-IV twice to 54 adults, three or six months apart, with no intervention of any kind: Full Scale IQ rose "approximately 7" points, Processing Speed about 9.

A single score is a band, not a number. A Wechsler IQ uses a scale with a standard deviation of 15, so the standard error of measurement is 15 × √(1 − reliability): about 2.1 points at a reliability of .98, about 3.4 at .95. The error on a difference between two scores is larger by a factor of about √2. Duff's 2012 review in Archives of Clinical Neuropsychology gives the formulas and the convention that a change counts as reliable at roughly ±1.645 standard errors of the difference — here, about ±5 to ±8 IQ points. We did not buy the Wechsler technical manuals, so those reliability figures are the published range rather than numbers read in a manual; the conclusion does not turn on which end you pick.

Scores move on their own, a lot. Ryan, Glass and Bartels retested 43 schoolchildren on the WISC-IV about eleven months apart with no intervention: "Mean practice effects were not significant, but range of gain or loss for some individuals was large. On the FSIQ, 42% changed > +/-5 points on retest. The FSIQ is less stable than one might infer from the large stability coefficient and small mean practice effect."

And the largest real effect available is small. Roberts and colleagues, in European Neuropsychopharmacology (2020), meta-analysed the three best-studied pharmacological cognitive enhancers in healthy, non-sleep-deprived adults across 47 studies. Modafinil: standardised mean difference 0.12 overall. Methylphenidate: 0.21. D-amphetamine: no effect. On a 15-point scale, roughly 2 and 3 points. Individual sub-domains run higher — methylphenidate reached 0.43 on recall and 0.42 on sustained attention, about 6.5 and 6.3 points, and those are the largest figures in the paper. Their conclusion: "There is a user perception that these drugs are effective cognitive enhancers, but this is not supported by the evidence so far."

Put the four together. Sitting the test again is worth about 7 points. The reliable-change band is about ±5 to ±8. Forty-two per cent of people move more than five points with nothing done to them. The best overall drug effect anyone has demonstrated is 2 to 3 points, and the largest sub-domain effect in that literature is about 6.5. Not one of those numbers is bigger than the practice effect, and every one of them is smaller than the ±8 upper end of the reliable-change band. A personal before-and-after cannot separate signal from noise — which is as true of a perfectly administered clinical WAIS-IV as of anything sold online.

Consumer IQ sites are where the experiment usually gets attempted. IQ Revealed publishes one: read on 2026-09-05, its homepage described a 40-question matrix-reasoning test taking about 15 minutes and returning "your IQ score, its classification band and your percentile ranking", with the statement "Results are for educational and entertainment purposes only and are not a clinical evaluation." Whatever the test, the arithmetic above is what governs. A single score is a rough placement, not an instrument sensitive enough to find a two-to-three-point drug effect through a seven-point practice effect.

Detecting an effect of that size would need randomisation, a placebo arm, blinding, alternate forms and enough participants to push the standard error of the group mean below the effect. That is a clinical trial. For four of the six compounds here, nobody has ever run one.

What this page does not do

It does not say what to take, how much, how often or by what route, and it contains no protocol. It does not rank the compounds or the vendors. It does not say any of these substances is safe, unsafe, legal or illegal in a particular circumstance — the safety statements here are quotations from regulators and reviewers, the legal ones from statutes and letters. It reports what the documents say, links each one, and marks where the documents say nothing.

Sources, and what we could not open

Opened 2026-09-05 unless otherwise dated: FDA's semax briefing document and the PCAC meeting page for 23-24 July 2026; FDA's Category 2 list and its sub-list of withdrawn nominations (page current 2026-04-22); FDA's 503A bulk drug substances page; FDA warning letters to The Guyer Institute of Molecular Medicine (2021-11-10, closed out 2026-02-24) and Crystal Clear Supplements (2022-02-04); 21 U.S.C. 353a and 21 CFR 10.20 on Cornell LII; the JPET retraction notices of April 2025 and the 2021 Notices of Concern; the DOJ release on the Athira settlement (2025-01); Athira's LIFT-AD topline release (2024-09-03); Cochrane's summary of CD007026.pub7 (2023-10-11); the CASTA abstract in Stroke (2012); the ADDF Cognitive Vitality report on dihexa (updated 2021-08-13); the WSU press release of 2012-10-11; abstracts for Calamia 2012, Hausknecht 2007, Estevis 2012, Ryan 2010 and Roberts 2020 via Europe PMC; ClinicalTrials.gov and Europe PMC search counts; RAPS and Mintz on the July 2026 vote; and the vendor pages in the claims table.

Could not be opened on the day, and therefore not quoted: the Health Affairs Forefront commentary on the vote, the Drug Topics meeting preview, the Taylor & Francis pages for Calamia et al. and Estevis et al. (abstracts read via Europe PMC instead), PubMed's own interface, and the Cornell eCommons copy of Hausknecht et al. The Wechsler technical manuals sit behind a publisher paywall we did not buy, so the reliability figures in the measurement section are the published range and are labelled as such. Washington State University's revocation of a doctorate is stated here as reported, not verified.

Corrections, with a document, go to the contact page; who publishes this site is on the about page.