Cerebrolysin (FPF 1070)

What is actually in the ampoule

Cerebrolysin is not registered as a chemical. The FDA Global Substance Registration System files it under substance class structurallyDiverse, with source material class ORGANISM, source material type MAMMAL, part brain, fraction protein, and parent substance SUS SCROFA WHOLE. The record carries UNII 37KZM6S21G and CAS registry number 12656-61-0. PubChem returns no compound identifier for the name at all, only thirteen substance deposits. There is no formula to check, no molecular weight and no sequence, because the preparation is a hydrolysate rather than a defined molecule. Every number that follows describes a batch of a biological mixture, and the identity fields on this page are null because the registries hold no value to put in them. The GSRS name list also carries marketed-product designations, which this page omits.

Two incompatible descriptions of that mixture circulate inside the peer-reviewed literature, not merely on secondary pages. Hartbauer and colleagues, writing in the Journal of Neural Transmission in 2001, described a mixture of 75 per cent free amino acids and 25 per cent low-molecular-weight peptides below 10 kDa, and their companion paper adds the qualification that the split is calculated from total nitrogen content. Allam and colleagues, in The Spine Journal in 2018, described 85 per cent free amino acids and 15 per cent biologically active low-molecular-weight peptides. Neither cites a characterisation study. The 85/15 figure also appears in papers from 1992, 1996 and 1997, which suggests an older estimate that was never formally superseded.

Mass spectrometry has since been pointed at the peptide fraction, and what it returned is broader and less tidy than either split implies. Gromova and colleagues (2019), using orbital ion trap instruments and de novo sequencing, reported 14,635 peptide sequences corresponding to 1,643 porcine neuronal proteins, of which 405 fragments matched 300 known biologically active peptides. Yang and colleagues (2023), optimising a nanoLC-MS workflow, found peptides belonging to myelin basic protein and tubulin beta chain, and nominated three candidates with predicted neuroprotective function by bioinformatic analysis rather than by assay. Both are descriptive surveys of what the material contains, and neither tested whether any single constituent produces a clinical effect.

The stroke trials, and the Cochrane reading of them

CASTA is the largest randomised trial of the preparation. Heiss and colleagues (2012) randomised 1,070 patients with acute ischaemic hemispheric stroke within twelve hours of onset to 30 mL daily by intravenous infusion for ten days or saline, both on a background of 100 mg aspirin. The confirmatory endpoint, a combined global directional test of the modified Rankin Scale, Barthel Index and National Institutes of Health Stroke Scale, showed no significant difference between groups. A post hoc analysis restricted to patients with NIHSS above 12 reported 90-day cumulative mortality of 20.2 per cent on placebo against 10.5 per cent on treatment. The authors described that as a trend requiring confirmation in a further trial.

CARS took a different question. Muresanu and colleagues (2016) randomised 208 patients starting 24 to 72 hours after stroke onset, gave 30 mL daily for 21 days alongside a standardised rehabilitation programme, and measured the Action Research Arm Test at day 90, reporting a Mann-Whitney estimator of 0.71 (95 per cent CI 0.63 to 0.79). The trial was funded by EVER Neuro Pharma, three of its authors were employees of that company, and the published paper calls the study exploratory and asks for large-scale confirmation. A second identical trial, CARS-2, was run but has no standalone publication that this search could locate; its data appear only inside a 2017 pooled individual-patient analysis of both trials, N equals 442, Mann-Whitney 0.62.

The 2023 Cochrane review is the closest thing to a synthesis. Ziganshina and colleagues included seven randomised trials and 1,773 participants, one of which tested a related porcine-brain peptide preparation and contributed 272 people. All-cause death showed a risk ratio of 0.96 (0.65 to 1.41) across six trials and 1,689 participants at moderate certainty. Non-fatal serious adverse events went the other way: risk ratio 2.39 (1.10 to 5.23) across three trials and 1,335 participants, also moderate certainty, rising to 2.87 (1.24 to 6.69) in the subgroup receiving 30 mL for ten days. Total serious adverse events showed no difference (RR 1.16, 0.81 to 1.66), as did fatal serious adverse events (RR 0.90, 0.59 to 1.38); the increase sat in the non-fatal subcomponent. The review records that no included study reported poor functional outcome defined as death or dependence, and that the manufacturer supported three of the multicentre studies wholly or in part.

Observational work has continued alongside. C-REGS2 (Vosko and colleagues, International Journal of Stroke, 2025, PMID 40851188) enrolled 1,865 patients across sixteen countries between April 2018 and April 2024, restricted to moderate stroke with a baseline NIHSS of 8 to 15, and reported a Mann-Whitney estimator of 0.6157 (95 per cent CI 0.5910 to 0.6404) on the 90-day modified Rankin Scale in a target population of 1,769, of whom 1,021 were treated and 748 were controls. It is an open-label comparative effectiveness study using cohort standardisation, not a randomised trial, and its authors present it as such. The ClinicalTrials.gov record, NCT03480698, posts enrolment as 1,851 against the 1,865 reported in the paper.

Dementia: two Cochrane reviews and a dose response that runs backwards

Cui and colleagues updated the Cochrane review on vascular dementia in 2019 and found nothing new to add. Six randomised trials with 597 participants were eligible, all of them already present in the 2013 version, and no trial published between 2012 and May 2019 met inclusion criteria. Pooling Mini-Mental State Examination and ADAS-cog+ data across three studies and 420 people gave a standardised mean difference of 0.36 (0.13 to 0.58) at very low quality. Global function response across two studies and 379 participants gave a risk ratio of 2.69 (1.82 to 3.98), again very low quality. Where funding was described, every study was industry-supported. The review authors wrote that if benefits exist they may be too small to be clinically meaningful.

The largest single vascular dementia trial sits inside that pool. Guekht and colleagues (2011) randomised 242 patients to 20 mL daily as add-on to acetylsalicylic acid over two treatment cycles, and reported ADAS-cog+ improvement of 10.6 points against 4.4 on placebo, which the paper states as a least-squares mean difference of -6.17; on ADAS-cog+ lower scores indicate less cognitive impairment, so the negative sign carries the direction of the estimate. CIBIC+ favoured treatment by 0.84 points at week 24, a mean of 2.84 against 3.68. Two of the four lead authors were employees of the sponsoring manufacturer.

In Alzheimer's disease the record is older and the dose behaviour is the interesting part. Bae and colleagues (2000) randomised 53 patients in Korea over four weeks and reported ADAS-Cog and CGI differences. Ruether and colleagues (2001) randomised 149 patients to 30 mL or placebo in two four-week courses separated by a two-month interval and reported a 3.2-point ADAS-cog difference at week 16 that was largely maintained at week 28; that paper carries an erratum in the following issue. Alvarez and colleagues (2006) randomised 279 patients across 10 mL, 30 mL, 60 mL and placebo arms, and at week 24 only the 10 mL arm improved significantly on ADAS-cog, while the two higher doses improved global impression but not cognition. The authors named the pattern a reversed U-shaped dose-response; they proposed no mechanism for it.

A 2025 systematic review in Alzheimer's and Dementia placed these results in a wider frame. Masserini and colleagues extracted 173 trials covering 22,347 participants and 91 interventions in vascular cognitive impairment (PMID 41198594). In that review the pooled estimates placed cerebrolysin in a group with acetylcholinesterase inhibitors, memantine, propentofylline, physical exercise and cognitive rehabilitation that showed small to moderate cognitive improvement; the largest estimates in the review were for Ginkgo biloba extracts, at Cohen's d 0.83 (95 per cent CI 0.00 to 1.67) for cognition and 0.50 (0.25 to 0.75) for function, and the authors record certainty of evidence as overall low. Their stated conclusion concerns the field as a whole: methodological heterogeneity across these trials undermined the strength of evidence for positive and negative findings alike.

Traumatic brain injury: a terminated trial and a single-centre successor

CAPTAIN I is registered as NCT01606111 and the registry lists it as terminated, with the reason given as poor patient recruitment. Forty-six patients were enrolled, 22 to treatment and 24 to placebo, against a design built around a multidimensional ensemble of fourteen outcome scales analysed by the Wei-Lachin procedure. In the intention-to-treat population the primary ensemble missed statistical significance. The per-protocol analysis reached it, with a combined Mann-Whitney of 0.69 (0.53 to 0.85). Three individual outcomes reached stand-alone significance. The paper carries an erratum published in March 2020.

CAPTAIN II was single-centre. Muresanu and colleagues (2020) enrolled 142 patients with admission Glasgow Coma Score between 7 and 12, analysed 139, and reported that the primary ensemble of thirteen outcome scales favoured treatment at day 90 with a combined Mann-Whitney of 0.59 (0.52 to 0.66) and p equal to 0.0119, an effect the authors characterised as small to medium. Safety observations were reported as comparable between groups.

Read together, the TBI file is a 46-patient trial stopped for recruitment whose primary endpoint failed on intention-to-treat, and a 139-patient single-centre trial with a small-to-medium ensemble effect. Neither is a confirmatory trial, and both papers say so in their own conclusions.

Four retractions and an expression of concern in the preclinical record

The transgenic-mouse literature behind the mechanistic vocabulary that surrounds this preparation came from one laboratory at the University of California San Diego, with co-authors employed by the manufacturer. Since March 2025, four of those papers have been retracted: the Rett syndrome study in Acta Neuropathologica (2008, retracted March 2026), the CNTF-peptide neurogenesis comparison in the Journal of Alzheimer's Disease (2011, retracted 2026), the tauopathy mitochondrial-structure paper in BMC Neuroscience (2014, retracted December 2025), and the three-repeat tau Pick's disease paper in BMC Neuroscience (2015, retracted March 2025). A fifth, the 2009 AAV2-mutant tau paper in Acta Neuropathologica, received an editorial expression of concern in March 2026.

The retraction notices are specific about what was found. For the 2015 Pick's disease paper, the editor recorded that a portion of one dentate panel appeared to overlap, when horizontally flipped, with a portion of another panel in the same figure, together with four further overlaps across two more figures, every one of the five described as panels representing animals that underwent different treatments or were subject to different experimental conditions. For the 2014 mitochondrial paper, the notice lists apparent overlap within a phospho-tau band, apparent editing of the background of an actin band, apparent overlap among three panels representing different experimental conditions, and apparent overlap within a phospho-Drp1 band. That notice adds that the corresponding author has not stated whether he agrees or disagrees with the retraction. Employees of EVER Neuro Pharma are listed as co-authors on both.

A separate cluster carries three more retractions: two 2018 Molecular Neurobiology papers on nanowired delivery of the preparation in Parkinson's and Alzheimer's models, and a 2023 Neurochemical Research paper on nanoparticle-induced neuropathic pain. None of the human trials cited on this page carries a retraction, expression of concern or unresolved erratum. The flags sit entirely on the animal and cell work, which is where the mechanistic vocabulary that surrounds this preparation was generated. The 2012 pharmacology review that states in its second sentence that the preparation mimics endogenous neurotrophic factors (PMID 22514792) was co-authored by the same corresponding author and draws on that same body of experiments.

Regulatory position and the missing pharmacology

There is no United States approval. Queries to the openFDA Drugs@FDA endpoint and the structured product labelling endpoint both return no matches for the name, which is what would be expected for a substance never approved and never labelled for the US market. The GSRS record carries FDA orphan drug code 513315. The FDA Office of Orphan Products Development record for that code, designated 5 April 2016 with EVER Neuro Pharma GmbH as sponsor, gives the designation as treatment of frontotemporal dementia, including all subvariants, and gives orphan approval status as not FDA approved for the orphan indication. The two registries disagree on current standing: the GSRS code comment reads 'Designated', while the OOPD listing reads 'Designated/Designation Withdrawn or Revoked'. The date of any withdrawal is not published in the listing.

ClinicalTrials.gov holds 42 studies naming the preparation. Ten list EVER Neuro Pharma as lead sponsor: one was withdrawn before enrolment, three were terminated, five completed, and one, a 30-participant study in CADASIL registered as NCT05755997, is active and not recruiting. The rest are investigator-initiated studies across stroke, traumatic brain injury, subarachnoid haemorrhage, cerebral palsy, Down syndrome, preterm neurodevelopment, delirium, Bell's palsy and post-COVID chemosensory dysfunction, most of them registered at 20 to 150 participants. The registered record therefore covers many indications at small scale.

What no search returned is a pharmacokinetic study in humans. No plasma concentration curve, no half-life, no brain-to-plasma ratio, no biodistribution work was located in PubMed or Europe PMC for the preparation or its FPF 1070 designation. The one trial found in people without neurological disease is Funke and colleagues (1998), which gave 48 healthy men 10, 30 or 50 mL or saline daily for ten days and measured EEG power ratio and short-term word recall around hyperventilation-induced ischaemia; the authors recorded signs they described as overdosage at the highest volume, together with a small but significant reduction in blood pressure.

Harm appears in the published record as individual case reports, without an incidence estimate. Trimmel and colleagues (2024) describe a fulminant anaphylactic reaction in an 85-year-old man with subacute stroke following intravenous administration, laboratory-confirmed, and note that published cases of anaphylaxis with this preparation are rare. That the preparation is derived from mammalian brain tissue is a fact of its manufacture; no published assessment of transmissible spongiform encephalopathy risk specific to it was located during this search.

What is not known

No human pharmacokinetic data exist for this preparation. Searches of PubMed and Europe PMC returned no plasma concentration curve, no half-life, no brain-to-plasma ratio and no biodistribution study in any species, which means the central mechanistic premise, that low-molecular-weight fragments reach brain tissue after intravenous infusion, has not been measured. Composition is a second gap: two incompatible amino-acid-to-peptide splits are in print without a characterisation study behind either, batch-to-batch variability has not been published, and the proteomic surveys that do exist describe what is present without testing what is active. The largest randomised trial, at 1,070 patients, was neutral on its confirmatory endpoint, and the positive results come from smaller trials, most of them manufacturer-sponsored, several with manufacturer employees as authors. Two Cochrane reviews reached opposite-sounding conclusions on different indications, and the vascular dementia review graded all of its evidence very low quality and found no eligible trial published after 2012. Trials of the CARS design have never been repeated at scale, and CARS-2 has no standalone publication that this search could locate. Dose response has not been resolved: the one four-arm dose-finding trial in Alzheimer's disease found its cognitive effect at the lowest volume tested and not at three or six times that volume. Long-term exposure has not been characterised in any indication, and the only trial located in people without neurological disease ran for ten days in 48 healthy men in 1998. Adverse-event data are thin: the 2023 Cochrane review found non-fatal serious adverse events roughly doubled at moderate certainty across three trials, while the composite estimates over those same trials were null for total serious adverse events, for fatal serious adverse events and for total adverse events, and no included trial reported death or dependence as a functional outcome at all. Anaphylaxis appears in case reports without an incidence estimate. The regulatory record is also unsettled at the margin: the FDA holds an orphan designation for the substance in frontotemporal dementia, and the two federal registries that carry it disagree on whether that designation still stands, with no withdrawal date published. Finally, the preparation is manufactured from mammalian central nervous tissue, and no published assessment of transmissible spongiform encephalopathy risk specific to it was located during this search.

Questions