TAK-653

Ketamine works, and the reason it works appears to involve AMPA receptors. Blocking NMDA receptors on inhibitory interneurons releases a burst of glutamate onto pyramidal cells, that glutamate hits AMPA receptors, and the resulting signal drives BDNF release, mTOR activation and new synaptic connections [4][5]. If the AMPA step is what matters, then acting there directly should give the antidepressant effect without the dissociation, the abuse potential or the clinic visit.

That idea is decades old and has a graveyard attached to it. Drugs that agonise AMPA receptors outright cause seizures, because they activate receptors everywhere including ones that no neuron is currently signalling through. Positive allosteric modulators were the fix, but most of them retained enough intrinsic agonist activity to keep the seizure risk and — more surprisingly — to produce narrow, bell-shaped dose-response curves where a higher dose worked worse than a lower one [1]. Farampator's depression trial was stopped early over adverse events in a separate study, and Servier's tulrampator failed a 414-patient phase 2 as an SSRI adjunct [4].

TAK-653 was Takeda's attempt to strip out the agonism completely. It binds the AMPA receptor's ligand-binding domain only when glutamate is already there, and in cells it produces essentially no response without an agonist present [1]. The payoff was measurable: in rats, the exposure margin between the dose that improved recognition memory and the dose that caused convulsions was 419-fold on peak concentration and 1017-fold on exposure, against 42-fold and 122-fold for the comparator potentiator LY451646 and roughly 4-fold for AMPA itself [1].

In humans it has been through phase 1 and a pair of linked pharmacodynamic studies in the same 24 volunteers. A single 6 mg dose significantly increased motor-evoked potential amplitude to single-pulse transcranial magnetic stimulation — direct evidence that the drug reaches the brain and does what it is supposed to do there [2]. On a battery of psychomotor tests the profile looked like a mild stimulant: faster saccades, better smooth pursuit and adaptive tracking, and no change in body sway, alertness ratings, mood ratings or "feeling high" [3]. Compared with the reference drugs, the saccadic effect was larger than caffeine 60 mg or dexamphetamine 20 mg and smaller than modafinil 200 mg, while the tracking improvement was comparable to all three [3].

Then the file goes quiet in an unsatisfying way. Neurocrine licensed the compound and ran a phase 2 trial in major depressive disorder. According to the review that summarises it, osavampator taken orally once a day significantly improved MADRS scores against placebo at day 28 and at day 56 — but those results exist only in a company press release, and whether the day 7 outcome was positive, which would determine if it counts as a rapid-acting antidepressant at all, was not mentioned [4]. Two phase 3 trials as adjunctive treatment in treatment-resistant depression started in January and March 2025 [4].

The whole design rests on not being an agonist. TAK-653 binds the ligand-binding domain of recombinant AMPA receptors in a glutamate-dependent manner, and in GluA1-expressing cells it increases calcium influx only when glutamate is present, with an EC50 of 3.3 µM [1]. The structural explanation is steric: the compound's cyclohexyl ring clashes with Ser743 in GluA1 (Ser750 in GluA2) in the channel-closed state, so it cannot stabilise the receptor's active conformation on its own. Mutating that serine to alanine removes the clash and restores a large agonist-like response, which is about as direct a demonstration of a design hypothesis as medicinal chemistry produces [1].

In rat primary hippocampal neurons it augmented AMPA-elicited currents with an EC50 of 4.4 µM while showing little agonistic effect of its own [1]. The interesting comparison is at the synapse. In prefrontal cortical slices, with GABA-A, GABA-B and NMDA receptors pharmacologically blocked, 10 µM TAK-653 enhanced the suprathreshold response to stimulation of glutamatergic afferents more than either AMPA or LY451646 did [1]. A drug with no agonism outperformed drugs with agonism at amplifying real synaptic transmission — which is the paper's central claim, and the reason it argues that tonic activation of resting receptors actively interferes with phasic signalling rather than adding to it.

Downstream it runs the ketamine cascade. In rat primary cortical neurons TAK-653 increased phosphorylated mTOR and p70S6 kinase along with Akt and ERK, and raised BDNF protein [6]. In rat hippocampal neurons it increased BDNF protein in the presence of a low AMPA concentration, and in mice it raised hippocampal BDNF mRNA at 3 and 10 mg/kg orally [1][7]. The link to ketamine was made explicitly in the same paper: ketamine at 30 mg/kg intraperitoneally produced an antidepressant-like effect in the rat reduction of submissive behaviour model, and pretreatment with the AMPA antagonist NBQX abolished it — so AMPA activation is necessary for ketamine's effect in that model, and TAK-653 given sub-chronically for six days produced the same behavioural effect directly, without the hyperlocomotion that tracks ketamine's psychotomimetic properties [6].

The cognitive profile is broader than the comparators and not bell-shaped. In rats, TAK-653 improved novel object recognition at 0.03, 0.1 and 0.3 mg/kg orally, and reversed an MK-801-induced working memory deficit in the radial arm maze across 0.1, 0.3, 3 and 10 mg/kg — a hundred-fold dose range. LY451646 improved recognition memory but not working memory, and AMPA improved neither [1]. In monkeys, a single 0.06 mg/kg oral dose improved delayed match-to-sample accuracy at a 16-second delay, with the effect gone by 48 hours [1]. Attention in the rat five-choice serial reaction time task did not improve in the population as a whole [1].

Repeated dosing did not blunt the response. Downregulation or desensitisation after chronic stimulation is a standing worry for any receptor activator. After 14 days at 0.3 mg/kg orally in mice, the hippocampal BDNF and Gadd45b response to an AMPA challenge was preserved [1].

In humans, target engagement is confirmed and nothing more. Single-pulse TMS of the motor cortex is a crude but honest readout of cortical excitability, and 6 mg raised motor-evoked potential amplitude against placebo at 2.5 hours, with the low 0.5 mg dose doing nothing [2]. The same doses in rats had tracked plasma concentrations: no effect at 0.1 mg/kg, an increase at 0.3 mg/kg and above, corresponding to 5.32 ng/mL or higher [2]. That is a clean translational bridge, and the authors are careful to note what it does not show — the motor cortex is not where mood is regulated, so an effect on corticospinal excitability says the drug engages its target, not that it will treat depression [2].

Across the two healthy-volunteer studies reported here, TAK-653 was well tolerated. All drug-related adverse events were mild, there were no serious adverse events and no withdrawals for an adverse event. Treatment-emergent events occurred in 37.5% of participants at 0.5 mg and 50.0% at 6 mg against 29.2% on placebo, and the most frequent were somnolence (12.5% at each dose), headache (4.2% and 16.7%) and nasopharyngitis; somnolence and headache also occurred on placebo [2]. There were no clinically significant changes in vital signs, ECGs or laboratory measurements [3]. In the nine-day repeat-dose interaction study, all adverse events were mild or moderate, somnolence was reported by half the participants in one arm, headache by about a third in another, and the two discontinuations were for COVID-19 and tonsillitis, both judged unrelated [8].

Seizure risk is the defining safety question for this drug class, and it has been taken seriously. No seizures, convulsions, dissociative effects or euphoria were observed in the human studies, including in participants who also received transcranial magnetic stimulation, which itself carries a small convulsion risk [2][3]. In rats, convulsions appeared in a single animal at 100 mg/kg orally, giving exposure margins over the cognition-improving dose of 419-fold on plasma Cmax and 1017-fold on AUC — roughly ten times the margin of LY451646 and a hundred times that of AMPA itself [1]. The 6 mg ceiling in the human crossover study was nonetheless set to keep peak concentrations well below those at which partial seizures were seen in primates [3].

Drug interactions look unremarkable. In vitro data suggested the compound might induce CYP3A4, but a dedicated phase 1 study found it did not meet the regulatory definition of a CYP3A inducer: exposures of midazolam, ethinyl estradiol and levonorgestrel were essentially unchanged, so combined oral contraceptives and other CYP3A substrates — including SSRIs and statins — can be given alongside it without dose modification [8].

None of this is long-term safety. The longest human exposure described in these references is a couple of weeks, and the phase 2 and phase 3 safety datasets have not been published.