The most expensive item on this shelf is usually N-Acetyl Semax Amidate. PubMed indexes no papers on that molecule at all.
That statement is worth being precise about, because the two papers usually cited for it are real papers about something else. Both studied N-acetyl Semax with a free C-terminus, not the amidated form. One measured how fast it breaks down in biological media (Dokl Biol Sci 2013, PMID 23652441). The other mapped copper and zinc binding and ran a cell viability assay (Magrì et al., J Inorg Biochem 2016, PMID 27586814). A PubMed search pairing semax with amide returns no relevant records. So the compound being sold at a premium has no behavioural study, no cognition endpoint, no human dose, and no chemistry paper of its own.
That is the first thing worth knowing about this category. It is a much smaller research field than the marketing implies, and most of the important primary literature can be read in a couple of evenings.
The upgrade nobody tested
The pharmacokinetic argument for acetylation is reasonable on paper. Cap the ends of the peptide and enzymes chew through it more slowly. If that were the whole story, acetylation would be a durability change rather than a new mechanism.
It isn’t the whole story, and the one published biological comparison went the wrong way. Magrì 2016 compared acetylated Semax against plain Semax in SH-SY5Y human neuroblastoma cells. Acetylating the N-terminus flipped the copper complex from a CuN4 chromophore to a distorted CuN3O chromophore with a more positive redox potential, and turned a redox-stable complex into one that reacts with ascorbic acid. Plain Semax protected the cells against copper-induced toxicity. The acetylated version didn’t. The authors concluded that the free NH2 terminus plays a crucial role, which is exactly the group acetylation removes.
So acetylation removes the copper-coordinating group that the protective effect appeared to depend on. That is a change to the mechanism rather than to durability alone, and it went in the unfavourable direction in the only experiment that looked.
N-Acetyl Selank Amidate is in a worse position. Searches for that molecule return only papers on plain Selank. No synthesis paper, no stability study, no cell assay. Every claim attached to it is borrowed from a different compound.
Where this research comes from
Nearly all of the Semax and Selank literature traces to one institutional cluster in Moscow, mainly the Institute of Molecular Genetics and the Kurchatov Institute with Myasoedov as a recurring senior author, plus the Zakusov Institute of Pharmacology. Much of it is Russian-language and abstract-only, with no effect sizes and no p-values reported. Independent replication outside Russia is close to nonexistent.
There are no registrations for semax, selank or dihexa on ClinicalTrials.gov. The often-cited n=110 post-stroke rehabilitation study (PMID 29798983) wasn’t placebo-controlled and reports no statistics in its abstract. Selank’s Russian approval rests largely on Zozulia 2008 (PMID 18454096), where 30 patients with generalised anxiety disorder and neurasthenia received selank and 32 received medazepam, a benzodiazepine. The anxiolytic effects were described as similar, with selank also showing antiasthenic and psychostimulant effects. The abstract gives no scale deltas, no p-values and no responder rates, so the comparison is qualitative as published. A second human selank trial exists and is rarely cited: Medvedev 2015 (PMID 26356395), n=70, selank plus phenazepam against phenazepam monotherapy, which administered Stroop and verbal fluency testing alongside HDRS, CGI, Spielberger, UKU and SF-36. Its abstract is equally free of statistics.
On healthy cognition specifically, the accurate claim is narrower than the one usually made in either direction. No verifiable, controlled study of Semax or Selank measuring cognitive performance in healthy people is retrievable. A 15-year program review from the originating group (Ashmarin 1997, PMID 9173745) states that Semax significantly improves memory and attention in healthy men under extreme conditions at 0.015 to 0.050 mg/kg intranasal, but the underlying work isn’t indexed and the review reports no design details, so it can’t be checked.
The two retrievable healthy-volunteer studies are both resting-state fMRI (PMID 30225715, n=24, and PMID 32342318, n=52). Lebedeva 2018 scanned 14 people on intranasal 1% Semax against 10 controls at baseline, five minutes and twenty minutes, and found greater volume of the rostral default mode network subcomponent in the Semax group. No effect size or p-value appears in the abstract, and no memory test, attention task or reaction-time measure was administered. It is a brain-imaging signal rather than a cognition result, though it is placebo-controlled, which makes the sweeping claim that no controlled Semax trial exists in PubMed wrong as usually stated.
What the BDNF data actually shows
The standard vendor line is that these peptides raise BDNF. The primary data say the direction depends on region, timing and baseline state.
Dolotov 2006 (PMID 16996037) is where the claim originates. A single 50 µg/kg intranasal dose in male Wistar rats produced a maximal 1.4-fold increase in hippocampal BDNF protein, a 1.6-fold increase in trkB phosphorylation, a 3-fold increase in exon III BDNF mRNA and a 2-fold increase in trkB mRNA, alongside what the authors called a distinct increase in conditioned avoidance reactions with no number attached.
Shadrina 2010 (PMID 19662538) sampled the same system at 20 minutes, 40 minutes, 90 minutes, 3 hours, 8 hours and 24 hours across hippocampus, frontal cortex and retina. Expression of both neurotrophin genes went down in hippocampus and retina at 20 minutes while going up in frontal cortex, and retinal BDNF rose only by 90 minutes. The authors describe the effect as multidirectional, in those words.
Kolik 2019 (PMID 31625062) gave outbred rats 10% ethanol as their only fluid for 30 weeks, then 0.3 mg/kg/day of selank intraperitoneally for seven days. Two separate results came out of it. In animals not exposed to ethanol, selank produced a cognitive-stimulating effect (p<0.05). In ethanol-exposed animals, it blunted withdrawal-related memory and attention disturbances (p<0.01) and prevented the ethanol-induced BDNF rise in hippocampus and frontal cortex (p<0.05). Those are different animals and different comparisons, and the abstract doesn’t link the behavioural benefit to the BDNF change, so the tempting reading that one caused the other isn’t supported by what was reported. Filatova 2017 (PMID 28293190) adds that selank alone changed none of 84 GABAergic and neurotransmission genes in IMR-32 cells, and when co-applied with GABA it largely suppressed GABA’s own gene expression changes.
Regional and temporal heterogeneity is a fair reading of all this. The stronger interpretation, that these compounds are homeostatic modulators pushing a system toward a set point rather than levers adding more growth factor, is a hypothesis that fits the pattern rather than something any of these studies tested. It spans three incommensurable designs: BDNF protein plus mRNA in rat hippocampus at one timepoint, mRNA only across three tissues and six timepoints, and hippocampal and cortical BDNF protein in a 30-week ethanol model using a different peptide.
Route changes the pharmacology
Shevchenko 2006 (PMID 16523722) gave outbred rats a single intranasal 50 µg/kg dose of tritium-labelled Semax at 56 Ci/mmol in 20 µL. At two minutes, brain contained 0.093% of the administered radioactivity per gram, and only about 80% of that was intact Semax. The paper reports no intravenous or intraperitoneal comparator, so that figure doesn’t support a claim about relative efficiency in either direction, and the authors’ own framing is that labelled Semax reaches the brain rapidly. The finding that does carry weight is the metabolite profile: across biological samples, the tripeptide Pro-Gly-Pro dominated over Semax and every other metabolite. How much of the reported activity belongs to Semax rather than to PGP, which the same group studies as an active compound in its own right, is genuinely unresolved.
Vasil’eva 2016 (PMID 29787664) is the study that settles whether route is a convenience choice. Mice received 300 µg/kg/day of selank for five days, either intraperitoneally or intranasally. Intraperitoneal selank raised frontal cortex GABA receptor binding sites by 38% with no change in hippocampal NMDA binding, while intranasal selank raised hippocampal NMDA binding sites by 23% with no effect on GABA receptors. Same molecule, same dose, different pharmacology.
The same paper found the behavioural effects appeared only in BALB/c mice, the high-anxiety, low-exploration strain, and not in normal-baseline C57BL/6 mice. Baseline state decided whether anything happened at all.
The two with the biggest claims and the worst paper trail
Dihexa’s HGF/c-Met mechanism story came from Benoist et al. 2014 (J Pharmacol Exp Ther 351:390-402). That paper was retracted on 29 April 2025 (PMID 40312093). The notice states that following a Washington State University investigation, figures and erratum data were found to contain falsified or fabricated data, with two named authors solely responsible. The foundational 2013 behavioural paper (McCoy, PMID 23055539) separately carries an Expression of Concern issued in September 2021. Any page describing dihexa as an HGF/c-Met synaptogenic agent orders of magnitude stronger than BDNF is citing work that is retracted or formally flagged.
The record isn’t uniformly dead, and it is worth saying so. An independent group at China Pharmaceutical University (Sun et al., Brain Sci 2021, PMID 34827486) reported oral dihexa rescuing Morris water maze performance in APP/PS1 mice, attributed to PI3K/AKT rather than HGF/c-Met. That is one independent positive result in a mouse disease model, working through a different proposed pathway than the retracted paper claimed.
Dihexa has never been given to a human in any published study. The nearest thing is fosgonimeton, the dihexa-derived HGF/MET modulator that went into the LIFT-AD trial (NCT04488419, PMID 41393340). It failed the primary endpoint and every secondary one. The between-group difference to week 26 was −0.08 (SE 0.10) on the global statistical test, p=0.70, with ADAS-Cog11 at −0.70 (0.77), p=0.35 and ADCS-ADL23 at +0.67 (0.92), p=0.61. Discontinuation for adverse events ran 14.2% on drug against 4.6% on placebo, mostly injection site reactions. A separate rat study (Wells 2024, PMID 38489193, n=40 Wistar) reported dihexa failing on weight, motor and spatial learning measures, in a 3-nitropropionic acid Huntington’s model rather than a cognition-enhancement model.
Cerebrolysin is the opposite case, with dozens of registered trials. The 2023 Cochrane review of acute ischaemic stroke (PMID 37818733, 7 randomised trials, 1773 participants) found no effect on all-cause death, RR 0.96 (95% CI 0.65 to 1.41), and an increase in non-fatal serious adverse events, RR 2.39 (95% CI 1.10 to 5.23), rising to RR 2.87 (1.24 to 6.69) in the 30 mL for 10 days subgroup. None of the included studies reported poor functional outcome, quality of life, or time to restoration of capacity for work, and three of the multicentre studies were manufacturer-supported.
The dementia evidence is more mixed than either side usually presents. The industry-linked meta-analysis (Gauthier 2015, PMID 25832905) reported cognition SMD −0.40 at four weeks that was no longer significant at six months (SMD −0.37, 95% CI −0.90 to 0.16, p=0.171), but also reported global clinical change significant at both timepoints, with an odds ratio at six months of 4.98 (95% CI 1.37 to 18.13, p=0.0150), and global benefit significant at six months (MW 0.57, p=0.0010). One of three endpoints lost significance rather than the signal disappearing. The independent Glasgow meta-analysis (Alsulaimani and Quinn 2021, PMID 36324709, 8 trials, 793 participants) found cognition SMD −0.16, 95% CI −0.30 to −0.03, which excludes the null, and global function OR 2.64 (1.17 to 5.98). Those authors describe it as potentially beneficial in improving cognition while concluding the effect is probably less than would be considered clinically relevant. That same review found no difference in serious adverse events, RR 0.96 (0.78 to 1.18), which sits against the Cochrane safety signal rather than alongside it. It also carries a 2022 erratum, which is exactly the kind of notice the closing advice in this article says to check for.
What the published studies dosed
Everything below records what published studies administered. None of it is a recommendation.
Semax
Rodent work used 50 µg/kg as a single intranasal dose in the pharmacokinetic and BDNF studies. Human protocols vary widely. The post-stroke study used 6000 µg/day for 10 days, repeated as two courses separated by 20 days. The healthy-volunteer fMRI study used a single administration of a 1% solution, while ophthalmology protocols use a 0.1% solution by endonasal electrophoresis, a tenfold difference in preparation strength between human studies, which indicates no dose has been settled. Route is intranasal in essentially all work, with some rodent studies using intraperitoneal. No oral bioavailability data for Semax exists.
Selank
300 µg/kg is the rodent workhorse dose, used as a single dose for gene expression (PMID 26924987), as 300 µg/kg/day for five days for behaviour and receptor binding, during consolidation for memory retention (PMID 20919548), and as 0.3 mg/kg/day intraperitoneally for seven days in the ethanol study. Enkephalinase inhibition IC50 is 15 µM in human plasma (PMID 11550013). The human anxiety trial doesn’t state its dose in the English abstract, and the Russian registered product is a 0.15% nasal solution. Intranasal and intraperitoneal are not interchangeable, as covered above.
The acetylated forms
No efficacy dose exists at any dose, in any species, for either one. Figures in circulation are extrapolated from plain Semax, and for N-Acetyl Selank Amidate there is no source literature to extrapolate from.
Dihexa
Rodent only. From McCoy 2013 in rats: oral gavage at 2 mg/kg fully reversed the scopolamine deficit and 1.25 mg/kg gave partial reversal, intraperitoneal worked at 0.25 and 0.50 mg/kg but not 0.05, and intracerebroventricular worked at 0.1 nmol in 2 µL artificial CSF, significant days 5 to 8 at p<0.05. In 24-month-old rats, oral 2 mg/kg daily improved performance on most test days (p<0.05). Mouse work also exists (PMID 34827486). No human dose has been published.
Cerebrolysin
Intravenous in every controlled trial. The six mild-to-moderate Alzheimer’s trials Gauthier pooled used 30 mL/day. The stroke literature is more varied: the Cochrane subgroups cover 30 mL for 10 days, 50 mL for 21 days, and 10 mL and 50 mL for 10 days, with CARS (PMID 28707130) running three weeks starting 24 to 72 hours post-stroke. The 30 mL for 10 days schedule is the one Cochrane specifically flagged for a higher rate of non-fatal serious adverse events. There is no studied at-home or intranasal protocol.
On timing and cycle length, the longest documented human exposure for Semax is two 10-day courses with a 20-day gap. There is no published human dose-response, chronic safety data, drug-interaction data or long-term follow-up for Semax, Selank or dihexa.
The safety signals the trials themselves flagged are the injection site reaction rate that drove fosgonimeton discontinuations, the non-fatal serious adverse event signal that worsened at higher cumulative cerebrolysin dosing, and a documented anaphylaxis case report for cerebrolysin (PMID 39055722). Product identity is a separate open question. A Belgian government laboratory paper in Drug Testing and Analysis (2020, PMID 31667971) reported Semax and Selank appearing in seized unlicensed pharmaceutical preparations, noted these peptides have not completed any clinical trials, and built an LC-MS/MS screening method precisely because they circulate online as lyophilised powders and nasal sprays with no quality control.
Where the evidence is thin, and what to read next
The summary is short. Semax has real mechanistic rodent work and no controlled human cognition data. Selank has two human trials, neither reporting statistics in its abstract, and rodent data showing the effect depends on strain and route. The acetylated forms have no literature of their own, and the nearest comparison went against the acetylated version. Dihexa has a retracted mechanism paper, a flagged behavioural paper, one independent positive mouse result through a different pathway, a negative result in a non-cognition model, and a derivative that failed a large human trial. Cerebrolysin has the most trials of any of them, a contested safety signal, and has never been studied for cognitive enhancement in a healthy person.
Five worth reading in full: Shevchenko 2006 (PMID 16523722) for the metabolite profile, Shadrina 2010 (PMID 19662538) for the multidirectional neurotrophin data, Vasil’eva 2016 (PMID 29787664) for route and baseline, the Benoist retraction notice (PMID 40312093), and the 2023 Cochrane cerebrolysin review (PMID 37818733) read next to Alsulaimani 2021 (PMID 36324709), since they disagree on safety. Magrì 2016 (PMID 27586814) is the one to check before paying a premium for an acetylated form.
One habit is worth more than any of it. When a product page cites a study, pull the PMID and check the top of the PubMed record for a retraction notice, an Expression of Concern or an erratum before accepting the claim. In this category that check changes the answer more often than expected, and it applies to the sceptical sources as readily as the promotional ones.
This article is a review of published research. It is not medical advice, and nothing here is a recommendation to use any compound. Several compounds discussed have no published human data and are sold for laboratory research use only.
