J147

Identity, and the distance from curcumin

PubChem returns one record for J147: CID 25229652, formula C18H17F3N2O2, molecular weight 350.3, InChIKey HYMZAYGFKNNHDN-SSDVNMTOSA-N, IUPAC name N-(2,4-dimethylphenyl)-2,2,2-trifluoro-N-[(E)-(3-methoxyphenyl)methylideneamino]acetamide. The FDA Global Substance Registration System holds the same structure under UNII Z41H3C5BT9 with preferred name J-147, the same formula and a molecular weight of 350.3357. Two CAS numbers attach to that record. GSRS lists 1807913-16-1 as the primary code and flags 1146963-51-0 as non-specific stereochemistry. The second is the number that appears on catalogue and aggregator listings almost without exception. Both resolve to the same PubChem compound. ChEMBL indexes the molecule as CHEMBL2387144 at a maximum phase of 1.

Descent from curcumin is a search history rather than a structural relationship. The programme started from curcumin and made hybrids with cyclohexyl-bisphenol A to give the pyrazole CNB-001, reported by Liu in 2008 (PMID 18208543); the hydroxyl groups were then removed to give CNB-023, first described in Table 1 of Chen 2011 (PMID 22194796); a deliberately promiscuous reaction between 2,4-dimethylphenyl hydrazine and m-anisaldehyde followed, and the most active product was picked off a thin-layer plate. What survived that sequence is a trifluoroacetyl hydrazide bridging a dimethylphenyl ring and a methoxyphenyl ring. The diketone is gone, the phenols are gone, and no fragment of the curcumin scaffold remains. Pages describing J147 as a turmeric extract or a form of curcumin are describing the provenance of an idea.

Nothing bearing this name is approved anywhere. Queries to openFDA against Drugs@FDA and against the labelling endpoint both returned NOT_FOUND for J147 and J-147 on 18 August 2026, and no approval in any other jurisdiction was located. The Salk Institute holds the patents; Abrexa Pharmaceuticals holds the licence, and the conflict-of-interest statement on Kepchia's 2021 paper in Aging records that the compound's originating scientist served as an unpaid adviser to that company.

The 2011 paper, and the statistics printed in its figure legends

Chen's 2011 report in PLoS One is the founding document, and its in vitro numbers are specific. Cell death after trophic factor withdrawal in primary rat embryonic cortical neurons was prevented with an EC50 of 25 nM. Extracellular amyloid toxicity in rat hippocampal neurons was blocked with an EC50 of about 200 nM. Across six assays the figure legend gives a range of 10 to 200 nM, at three or four replicates per point. Curcumin, run alongside, was described as marginally active in two assays and inactive in the rest at the concentrations tested, which included 10 micromolar in the glucose starvation assay where J147 was used at 20 nanomolar.

Rat behaviour was contracted out. Seven-week-old male Sprague-Dawley rats received 1, 2 or 5 mg/kg by oral gavage one hour before training in a novel object recognition test, twelve per group, with galantamine at 3 mg/kg intraperitoneally as positive control; the recognition index rose at all three doses, F(3,40) = 12.9. That F value corresponds to p below 0.00001 at those degrees of freedom, and the reported statistics are internally consistent.

The wild-type mouse panel is a different matter. Male C57BL/6J mice, twelve per group, were fed 200 ppm in chow and run through a Barnes maze, a Y maze and a novel object location test. The legends report novel object location at F(1,22) = 2.5 with an asterisk for p below 0.05, the Y maze at F(1,22) = 3.3 with p = 0.0032, and Barnes maze retention at F(1,22) = 2.6 with an asterisk for p below 0.01. At one and twenty-two degrees of freedom the critical value for p = 0.05 is 4.30. Those three F values correspond to p = 0.13, p = 0.083 and p = 0.12. The F statistics and the p values cannot both be right, and the published record does not establish which is wrong.

Two further mismatches sit in the same panel. The Methods state that Y maze data were analysed by Student's t-test, while the result is reported with an F statistic. The Barnes maze acquisition trend is given as F(1,22) = 2.0 with p = 0.057 in the body text and described in the figure legend as almost significant with the note that p was above 0.05 for both blocks. Nothing here was checked against the underlying data, which are not deposited.

Transgenic mice, and what the control groups were

APPswe/PS1ΔE9 line 85 mice carried the amyloid work. In the 2011 paper the transgenic animals were fed 200 ppm in a high-fat chow and assayed in the Morris water maze, acquisition F(2,14) = 8.3 and probe F(2,54) = 16.6, at six to nine mice per group. Soluble hippocampal Aβ 1-40 and Aβ 1-42 fell; the RIPA-insoluble fraction did not change. The paper states plainly that a wild-type-plus-drug arm was omitted because behavioural data in wild-type rodents already existed. Start age is given three ways: three months in the Methods, four months in the Results, and three to ten months in the figure legend, with the histology panel describing thirteen-month-old animals.

Prior's 2013 study is the one usually cited for reversal rather than prevention. Male and female transgenic mice were aged to twenty months, then fed 200 ppm, about 10 mg/kg per day, for three months. Escape latency in a two-day water maze, open-arm time on an elevated plus maze and freezing to context all separated from control; freezing to the cue on day three did not. Group sizes disagree between the two places they appear: the body text says eleven mice on control food and thirteen on drug, the figure legend says eleven and twelve. No wild-type group was included at all, and the paper says so, citing prior characterisation of the line.

Senescence-accelerated SAMP8 mice supplied the ageing arm. Currais fed seventeen male animals control diet and eighteen the same diet with 200 ppm from three to ten months of age, with fourteen young animals as a reference group. Six control mice and two treated mice died during the study. That imbalance is reported in the Methods and is not analysed as a survival outcome anywhere in the paper. Behavioural comparisons ran at twelve to sixteen per group and western blots at six.

Pharmacokinetics, all of it in mice, from one experiment

Every pharmacokinetic figure in circulation traces to a single contract-laboratory run reported in Prior 2013. After one oral dose of 20 mg/kg in mice, terminal half-life was calculated at 1.5 hours in plasma and 2.5 hours in brain, oral bioavailability at 28%, brain concentration at roughly 600 nM two hours after dosing, and the brain-to-blood ratio at approximately 0.5. Strain, sex and number of animals are not stated. The experiment has not been repeated in any published work, and no pharmacokinetic data exist in any other species.

Earlier the same numbers were unavailable. The 2011 paper described the molecule's pharmacokinetics as good and cited them as not shown. Metabolism was addressed separately by Chiruta in 2013, using human and mouse liver microsomes and mouse plasma, and the concern being tested was specific: a phenyl hydrazide could in principle yield aromatic amines or hydrazines. It did not. The oxidative metabolites that were found were themselves neuroprotective in the same assays, which leaves the exposure attributable to parent compound unresolved.

Toxicology is thin and mostly unpublished. Rats and mice given 2 g/kg orally showed no acute toxicity, a result attributed to a named contract laboratory with no data shown. hERG, CYP3A4 and Ames assays were reported negative on the same basis. Lapchak's 2013 paper is the one place a toxicology dataset was published in full: a CeeTox panel in the rat hepatoma line H4IIE returned a CTox value of 90 micromolar, with adverse effects on mitochondrial function only at high concentrations, and no genotoxicity with or without Aroclor-1254. The safety-window ratios quoted from that paper are two figures for two endpoints rather than one range: a neuroprotection ratio of 783 to 1500 and a separate neurotrophism ratio of 3600, both arithmetic performed on previously published EC50 values rather than a measurement made in that paper.

One target, identified once

Goldberg's 2018 paper in Aging Cell named the alpha-F1 subunit of mitochondrial ATP synthase, ATP5A, as the binding partner. Two orthogonal methods produced it. Drug affinity responsive target stability on HT22 lysates preserved a band whose most enriched protein by mass spectrometry was ATP5A, and affinity precipitation with a biotinylated analogue pulled the same protein from HT22 and mouse subventricular zone lysates, with binding competed away by excess unlabelled compound. ATP5A was the only protein reproducibly identified by both methods.

Functionally, the effect measured was partial and small. In isolated bovine heart mitochondria, ATP synthase activity was inhibited by 23.6 plus or minus 3.4% at saturation with an EC50 of 20 nM, F(9,20) = 40.72. Knockdown of ATP5A by siRNA reproduced the increases in mitochondrial membrane potential and superoxide production that the compound produced. No dissociation constant is reported anywhere in that paper, so the affinity implied by the phrase high-affinity target is not a measured quantity.

Independent confirmation has not appeared. A PubMed search for the compound combined with ATP5A returns five records: the Goldberg paper, two reviews and two in-silico modelling papers, all citing it rather than testing it. Kepchia's 2022 paper, from the same institute, states that no biomarker of target engagement had been identified. The lifespan result in that literature is a fly result: median lifespan in male Drosophila rose 9.5% at 0.1 micromolar and, at 2 micromolar, 12.8% according to the Results text and 12.5% according to the figure legend and the Discussion of the same paper, F(2,386) = 7.654. No lifespan experiment in a mammal has been published.

Work from outside the originating laboratory

Stroke work has been the most active line since 2020, and its most useful result is a negative one. Jin dosed adult male Wistar rats of 280 to 320 g intravenously at 1, 10 or 30 mg/kg two hours after the onset of occlusion. In the suture transient model, infarct volume and Bederson score improved at 10 and 30 mg/kg but not at 1 mg/kg, nine rats per group. In the embolic model neither infarct volume nor neurological deficit differed from vehicle, six per group. The combination arm showed something different: given with tPA at four hours, haemorrhage volume was lower than with tPA alone, at nine to fifteen rats per group. Two co-authors are employed by the sponsor company, the study was funded by that company, and the disclosure records the funder's involvement in the pharmacokinetic work and in revising the manuscript (PMID 35309561).

Other groups have reported in models unrelated to dementia. Jin's 2024 controlled cortical impact study in mice reported dose-dependent neurofunctional recovery and reduced endoplasmic reticulum stress markers after oral dosing begun one hour after injury (PMID 39245209). A group at Wenzhou Medical University reported antidepressant-like and anxiolytic-like behaviour in mice after three days of dosing at 10 mg/kg, with raised serotonin and noradrenaline in frontal cortex and hippocampus and inhibition of monoamine oxidase A (PMID 34023306); that paper is behind a subscription and its abstract states neither the route nor the group sizes, so the dose figure carries a species but not the route and sample size the sourcing standard asks for. He's 2025 paper reported microglial polarisation through CAMKK2 and AMPK in a lipopolysaccharide model (PMID 40712391). None of these is an Alzheimer's model, and none has been replicated by a second laboratory.

Oberman's 2024 study in Zucker rats is worth separating out because it reports a mixed result and a failed premise. Chronic treatment in food reduced open-field anxiety and protected short-term spatial memory, single intravenous dosing at the start of surgery improved long-term spatial memory, and neither produced an anti-inflammatory effect. The authors also record that surgery did not induce extensive post-operative cognitive dysfunction in this strain, which is the deficit the intervention was meant to address. An Abrexa author is on that paper too.

Two registered trials, and no reported results from either

NCT03838185 was a randomised, double-blind, placebo-controlled single-ascending-dose study run by Abrexa Pharmaceuticals at one site in Overland Park, Kansas. Sixty-four participants were dosed in cohorts of eight, six on drug and two on placebo, comprising healthy males aged 18 to 50 and healthy men and women aged 60 to 85. The primary outcome measures were treatment-emergent adverse events, electrocardiogram abnormalities, laboratory panels and physical and neurological examination changes from pre-dose to seven days after. It completed on 1 February 2020. No results are posted on ClinicalTrials.gov.

Nor has it been published. A PubMed search for the compound restricted to the clinical trial publication type returns zero records, and a search combining the compound with healthy volunteers returns zero. Six and a half years after the last participant was dosed, no adverse-event table, no pharmacokinetic curve and no tolerated dose from that study exists in the public record. Statements that the trial demonstrated safety and tolerability in humans are not traceable to a document reporting what happened.

The second registration moved indication and route. NCT07430917, titled JUMPSTART, is a Phase 2 randomised, placebo-controlled, adaptive study of an intravenous emulsion given as a single injection after mechanical thrombectomy in acute ischaemic stroke, 196 participants estimated, at one site in Austin, Texas. The registry lists two dose levels; neither has been studied in any published report. Its primary outcome is safety over 90 days. First posted 24 February 2026 with an estimated start of 20 April 2026, it carried a status of not yet recruiting when the registry was queried on 18 August 2026. There is no registered trial of this compound in Alzheimer's disease or in any cognitive indication.

What is not known

No human result of any kind has been published for this compound. The one completed trial dosed 64 healthy volunteers and has posted nothing in the six and a half years since, which means there is no published human pharmacokinetic curve, no adverse-event table, no maximum tolerated dose and no exposure figure in any person. Everything known about absorption and distribution comes from a single unreplicated mouse experiment whose group size, strain and sex were not reported. Chronic toxicology, carcinogenicity, reproductive and developmental toxicity and drug-interaction data are absent from the peer-reviewed literature entirely; the acute toxicity result at 2 g/kg is a contract-laboratory finding cited without data, and the one-line statement that offspring of treated pregnant mice were normal carries no group size. The molecular target has been identified once, by one laboratory, with no dissociation constant and no independent replication in the eight years since. Whether the ATP synthase interaction accounts for any behavioural result has not been tested by anyone. The rodent behavioural record rests on models of familial Alzheimer's disease and accelerated senescence, neither of which is sporadic Alzheimer's disease, and several of those experiments ran without a wild-type comparison group. Nothing has been studied in cognitively healthy people, and no trial in a cognitive indication is registered anywhere.

Questions