J-147

J147 came out of a drug-discovery programme at the Salk Institute that deliberately skipped the usual step of choosing a molecular target. Because age is the largest risk factor for Alzheimer's disease, the group built a panel of cell assays for the insults an ageing neuron faces — loss of growth-factor support, glutathione-depleting oxidative stress, energy failure and amyloid toxicity — and required a candidate to work in all of them before it went into animals [1]. The starting point was curcumin. Hybrids with cyclohexyl-bisphenol A gave CNB-001, which was more stable and more potent; stripping its hydroxyl groups gave CNB-023; and J147 was then isolated from a deliberately messy reaction between 2,4-dimethylphenylhydrazine and m-anisaldehyde, as the one HPLC peak active in all three primary assays [1]. Despite the family tree it is not a curcuminoid; a later paper put it plainly that the two are not chemically related [7].

In culture it protected neurons at 10–200 nM across six assays, where curcumin was weak or inactive [1]. In healthy young animals, single oral doses of 1–5 mg/kg improved object recognition in rats. Mice fed it did better in object-location and Y-maze tests, with only a trend in Barnes maze learning that became significant in a retention test six weeks later [1]. It helped long-term potentiation form in rat hippocampal slices; an analogue with the two nitrogens replaced by carbons did nothing, in slices or in behaviour [1].

In Alzheimer's-model APPswe/PS1ΔE9 mice it was tested two ways. Fed from 3 to 10 months of age, it prevented the water maze deficit, lowered soluble Aβ and restored synaptic proteins [1]. More demandingly, mice already 20 months old with advanced pathology were fed it for three months. Spatial and contextual memory recovered, soluble Aβ and BACE fell and NGF and BDNF rose, while plaque counts did not change [2]. In the same paper, against scopolamine-induced amnesia it matched donepezil for working memory and beat it for spatial memory, and the combination was best for fear memory [2].

Rapidly ageing SAMP8 mice fed J147 from 3 to 10 months kept their object recognition and reversal learning. They also avoided the rises in hippocampal Aβ1-40, tau phosphorylation, vascular inflammation and plasma acylcarnitines seen in untreated old mice [3]. Started later, from 9 to 13 months, it prevented disinhibition and the rise in plasma creatinine but did not improve Barnes maze reversal learning, and it changed only 34 hippocampal genes against 684 for a sister compound [8][9]. Very old wild-type mice fed it from 24 to 30 months had more dendritic spines and more dividing cells in the dentate gyrus and noticed smaller object movements; exploratory activity, in the authors' words, was negatively affected [10]. It also extended median lifespan in male fruit flies by roughly 10–13% [4][9].

Other groups have since tried it well beyond dementia. In rat stroke, 10–30 mg/kg IV shrank infarcts after a suture occlusion but did nothing on its own after an embolic clot; given with late tPA it reduced both infarct size and tPA-induced bleeding [7]. In mouse brain injury, daily oral doses improved recovery over 35 days [11]. In type 1 diabetic mice, 20 weeks of twice-daily gavage preserved nerve conduction velocity and blunted allodynia, and single doses relieved established allodynia within an hour [12]. Mouse antidepressant screens were positive after one to three days [13][14][15], as was an LPS model of sepsis-associated depression, where it also reduced deaths [16]. In obese Zucker rats after abdominal surgery, chronic dosing in food reduced anxiety and protected spatial memory, a single IV dose helped different measures, and neither affected inflammation [17].

What is missing is people. A phase 1 trial finished in February 2020 and its results have never appeared [5][6]. The developer has moved the compound into acute stroke, with an intravenous phase 2 trial that began recruiting in 2026 [18].

The target is ATP synthase, and the inhibition is deliberately incomplete. Three methods converged on the α-subunit of mitochondrial ATP synthase (ATP5A): drug affinity responsive target stability, pull-down with a biotin-tagged J147 that unlabelled drug competed off, and localisation of the tagged compound to mitochondria within ten minutes [4]. In isolated bovine heart mitochondria J147 inhibited the enzyme with an EC50 of 20 nM, but never by more than about a quarter, even at saturating concentrations [4]. That ceiling matters: complete blockers such as oligomycin are poisons. Knocking ATP5A down with siRNA, or over-expressing the enzyme's natural inhibitor IF1, reproduced J147's protection against amyloid, oxidative and energy-failure toxicity, and J147 added nothing on top of the knockdown [4].

Counter-intuitively, cells end up with more ATP. J147 raised mitochondrial membrane potential and mitochondrial superoxide, and lifted whole-cell ATP within 4–6 hours without changing glycolysis [4]. It also raised cytosolic calcium, and AMPK was activated through the calcium-sensing kinase CAMKK2 rather than through an energy shortage; the CAMKK2 inhibitor STO-609 blocked it [4]. Downstream, AMPK phosphorylated raptor to damp mTOR and phosphorylated ACC1, and in AMPK-knockout cells J147's protection was much weaker [4]. Modest ATP synthase inhibition feeding into AMPK and mTOR is a known lifespan mechanism from worms to mammals, which is why the same paper went on to test fly lifespan [4][19].

ACC1 links the target to metabolism and memory. Inhibiting ACC1 stops acetyl-CoA being spent on fatty acid synthesis. In neurons that raised acetyl-CoA, protected against oxidative death and increased acetylation of histone H3 at lysine 9, a mark linked to memory formation, which was also higher in treated mouse cortex [8]. In liver the same switch lowered plasma free fatty acids in three separate rodent studies, which the Salk group has proposed as a blood marker that the drug has reached its target [20].

Calcium handling is the other thread. J147 and ATP5A knockdown both dampened store-operated calcium entry after the endoplasmic reticulum was emptied, and J147 kept the expression of ion-transport genes in old SAMP8 hippocampus closer to that of younger mice [21].

The neurotrophic signature is downstream. J147 raises BDNF and NGF in rodent hippocampus and switches on BDNF-responsive proteins such as Homer-1, Egr3 and phosphorylated PSD95 [1][2]. It does not act like BDNF itself: it protected HT22 cells equally well with or without the BDNF receptor TrkB [1]. It is not an acetylcholinesterase or phosphodiesterase inhibitor, and unlike its parent compounds it does not directly stop amyloid aggregating [1].

Other groups describe other pathways. In microglia, anti-inflammatory reprogramming after LPS depended on CAMKK2/AMPK [22], and TLR4/NF-κB activation was reduced [16]. After brain injury, endoplasmic-reticulum stress markers fell alongside changes in AMPK and SREBP-1 [11]. In the stroke model, J147 with tPA lowered MMP-9, 15-lipoxygenase-1 and PAI in ischaemic endothelium and reduced platelet activation [7]. In melanocytes it activated ERK and degraded the pigment transcription factor MITF [23]. Most of these fit an AMPK-centred picture; none has been tested against the ATP5A mechanism directly.

Off-target activity is real. The inventors' contract screen at 10 µM found reproducible inhibition of only two targets: the dopamine transporter (IC50 0.649 µM) and MAO-B (IC50 1.88 µM) [2]. The reported mouse peak plasma level of about 0.6 µM after 20 mg/kg is in the same range as the transporter figure [12]. A Wenzhou group later reported MAO-A inhibition with higher cortical serotonin and noradrenaline in mice [15], and a sub-nanomolar affinity for the 5-HT1A receptor (Ki 0.42 nM) whose blockade abolished J147's antidepressant-like effect [14]. That 5-HT1A figure is hard to square with the original screen of over 60 CNS receptors and transporters, which found nothing at 10 µM [1][2], and it has not been replicated.

J147 is a single defined small molecule — an N-aryl trifluoroacetyl hydrazone of 350 Da with a cLogP of 4.5 and a polar surface area of 41.9, numbers chosen to suit a brain-penetrant oral drug [1]. It crystallises as the E isomer. The crystal conformation matches the lowest-energy one calculated in the gas phase, but the barriers between its main rotamers are low, and a different conformer appears to dominate in chloroform solution [24]. It is fat-soluble enough that animal studies dissolved it in corn oil, ethanol with Solutol HS15, or DMSO with polyethylene glycol [1][7][12], yet it stayed soluble up to 300 µM in serum-containing culture medium [25].

Oral bioavailability in mice was 28% [2], against under 1% for curcumin in rats [6]. The developer's clinical products are an oral dose in the phase 1 trial, whose placebo was corn oil, and an intravenous emulsion at 20 mg/mL for stroke [5][18].

There is no published human safety data. The phase 1 trial measured adverse events, ECGs and blood and urine chemistry for a week after single oral doses in young and elderly volunteers [5], but nothing from it has been posted or published [6].

The animal toxicology that exists is mostly reported in summary by the inventors, with the data available on request. Single doses of 2 g/kg caused no acute toxicity in rats or mice; hERG, CYP3A4 and Ames tests were negative; and mice fed 10 mg/kg/day through pregnancy had normal offspring [2]. An independent in-vitro panel in rat liver cells found cell death only at very high concentrations (TC50 293 µM). ATP, MTT reduction and glutathione fell at 210–228 µM and cell proliferation slowed at 20–100 µM, with no effect on caspase-3 or lipid peroxidation and no mutagenicity in two Ames strains up to 0.36 mM [25]. The panel's estimate of the sustained concentration that would be toxic in a 14-day rat study was 90 µM [25]. None of this is a formal repeat-dose, reproductive or carcinogenicity programme.

Two chemical concerns were examined directly. As a phenyl hydrazide, J147 could in principle break down to aromatic amines or hydrazines, which are carcinogenic; in human and mouse microsomes and mouse plasma it did not — metabolism oxidised the rings and left the hydrazone intact [27]. And because it inhibits the dopamine transporter, the inventors looked for addiction in mice and report finding none, though that study has not been published [2].

Over months of feeding, body weight did not differ from controls in SAMP8 mice [3][8]. In healthy mice, 20 weeks of dosing left glucose, insulin, HbA1c, nerve conduction and rotarod performance unchanged [12]. It did raise paw heat-withdrawal thresholds in healthy as well as diabetic mice, an analgesic effect on normal pain sensing [12]. The monoamine activity — dopamine transporter and MAO-B inhibition in vitro, MAO-A inhibition in mouse brain [2][15] — is the most plausible route to drug interactions, and none has been tested.