Ampakines & AMPA receptors

Both receptors bind glutamate and both are ion channels, but they do different jobs. AMPA receptors gate fast and do the routine work of passing excitatory signals along; NMDA receptors sit blocked by a magnesium ion at resting potential and only conduct when glutamate arrives and the cell is already depolarised and a co-agonist such as glycine or D-serine is present. That triple requirement is what makes NMDA a coincidence detector and AMPA the thing that provides the coincidence ^[3].

The reason AMPA became a drug target is what long-term potentiation turned out to be made of. Strengthening a synapse largely means inserting more AMPA receptors into the postsynaptic membrane, and weakening one means removing them; receptor trafficking is the currency of plasticity rather than a side effect of it ^[4]. A drug that made existing AMPA receptors work harder was therefore expected to lower the threshold for the same machinery that encodes memory. See NMDA, glutamate & memory for the other half of that story.

Two complications are worth carrying forward, because they explain a lot of later disappointment. First, subunit composition changes the receptor's behaviour: RNA editing at the GluA2 subunit determines whether the assembled channel passes calcium at all, so two receptors made of the same family of proteins can be functionally different devices. Second, native AMPA receptors do not sit alone. They travel with auxiliary subunits, the TARP family including stargazin and the cornichon proteins, which alter gating kinetics, trafficking and pharmacology ^[3]. A modulator characterised on a bare recombinant receptor in a cell line is not guaranteed to behave the same way on a real synapse, and that is a general hazard rather than a hypothetical one.

There are two separate ways an AMPA receptor current stops, and almost everything about ampakine pharmacology follows from keeping them apart. Deactivation is glutamate leaving: the transmitter unbinds and the channel closes. Desensitisation is glutamate staying bound while the channel closes anyway, which is a distinct conformational process. A modulator can slow either one, and which one it slows determines what the drug does to a synapse.

The structural answer arrived in 2002. The ligand-binding cores of AMPA receptors assemble into dimers, and desensitisation happens by rearrangement of the interface between the two halves of that dimer. Mutations or allosteric modulators that stabilise the interface reduce desensitisation; perturbations that destabilise it make desensitisation worse ^[5]. That single finding converted a kinetic curiosity into a druggable pocket.

Three years later, aniracetam and the ampakine CX614 were co-crystallised with the GluA2 ligand-binding domain. Both bind within the dimer interface at a common site sitting on the twofold axis of molecular symmetry, next to the hinge of the clamshell that closes when glutamate binds; point mutations at the contacting residues disrupted the drugs' function in patch-clamp recordings. The proposed action is that they stabilise the clamshell in its closed, glutamate-bound conformation and so slow deactivation ^[8]. This is about as concrete as mechanism gets in this field.

Older tool compounds anchor the same pharmacology from the other direction. Cyclothiazide potentiates AMPA receptors with absolute selectivity over kainate receptors, while concanavalin A does the exact opposite, potentiating kainate and barely touching AMPA ^[6]. That mirror-image pair is how the field established that AMPA and kainate receptors assemble independently and can be modulated independently.

The founding pharmacological observation for this class is older still and is worth reading closely. Aniracetam reversibly potentiated ionotropic quisqualate and AMPA responses in oocytes and hippocampal slices; it increased the conductance change without altering the receptor's affinity for agonist or the channel's ion selectivity, and it left kainate, NMDA and GABA responses alone ^[7]. Clean, specific, and effective only above 0.1 millimolar, a concentration nobody has shown an oral dose produces in a human brain.

Finally, the effect is not uniform across the brain. Ampakine potentiation of excitatory postsynaptic currents is substantially larger in CA1 pyramidal cells than in thalamic neurons or in hippocampal interneurons ^[2], so a compound that looks strong in the hippocampus is not necessarily doing much elsewhere.

The reason interest in this class outlived the acute-signalling story is that ampakines also change gene expression. Treating cultured rat entorhinal and hippocampal slices with the ampakine CX614 markedly and reversibly raised BDNF messenger RNA and protein, and a second, structurally unrelated ampakine, CX546, did the same. The effect was blocked by AMPA receptor antagonists but not by NMDA antagonists, and by reducing transmitter release, which places it squarely downstream of AMPA-mediated network activity. Intraperitoneal CX546 raised hippocampal BDNF messenger RNA in aged rats and middle-aged mice ^[9].

That result is what turned ampakines from a signalling tweak into a plausible disease-modifying idea: a small molecule that raises the brain's own growth factor output without injecting a protein. See BDNF & neuroplasticity for why that is harder than it sounds.

The caveat is inside the same paper and is routinely dropped when the finding gets repeated. Under prolonged infusion, BDNF messenger RNA peaked at 12 hours and returned to baseline by 48 hours, while nerve growth factor messenger RNA came back down even faster ^[9]. The authors stated the conclusion plainly: the response becomes refractory during continued ampakine exposure. Protein stayed elevated across the 48 hours, so the picture is not simply that the effect vanishes, but the transcriptional signal that the whole neurotrophic argument rests on switches itself off under exactly the chronic dosing schedule any real treatment would use.

That is a specific, testable prediction of tolerance built into the mechanism, and it belongs beside every claim that ampakines raise BDNF. See tolerance & dependence for the general shape of this problem.

Aniracetam is the accidental founder. It was made as a racetam, and the AMPA finding came afterwards, which is how a chemical family defined by a lactam ring became entangled with a mechanism it does not share. It also comes with a number people skip: an oral dose is extensively cleared on first pass, and most of what reaches the blood is the metabolite N-anisoyl-GABA rather than aniracetam itself, which sits awkwardly beside an in vitro threshold above 0.1 millimolar ^[7].

The CX series came out of Gary Lynch's laboratory at Irvine and was developed by Cortex Pharmaceuticals. CX-516, also called Ampalex, is the type II prototype and by far the most tested in humans. CX546 is the type I reference compound ^[2] and CX614 is the one that appears in both the structural and the neurotrophin work ^[8][9]; both are research tools rather than clinical candidates. CX717 was taken in a different direction entirely, toward respiratory rather than cognitive indications. Tulrampator, also known as S 47445 or CX1632, is the most recent of the line to reach clinical development.

Two large pharmaceutical companies ran their own. Organon made farampator, also called Org 24448 or CX691, and later Org 26576. Eli Lilly made LY451395, mibampator, and took it into Alzheimer's disease. Neither programme produced an approved drug.

The one unambiguous human success in the class is not a cognitive result at all. AMPA receptors drive the brainstem rhythm generator that sets breathing, and opioids suppress it. In sixteen healthy men given a single 1,500 mg oral dose of CX717, the respiratory rate fell by only 2.9% under a target alfentanil concentration, against 25.6% under placebo, with blood oxygenation and the hypercapnic ventilatory response similarly protected; and crucially, analgesia was not reduced in either an electrical or a heat-based pain model ^[15]. That is a clean, mechanistically motivated, positive controlled result, and it is about breathing.

The CX-516 story is worth telling in order, because it is a textbook illustration of how a promising pilot dissolves. In 2001 a small placebo-controlled study added CX516 to clozapine in patients with schizophrenia: six patients in a dose-finding arm and thirteen in a fixed-dose arm. It was well tolerated and produced moderate to large between-group effect sizes on measures of attention and memory. The authors concluded that ampakines held promise ^[10].

In 2008 the properly powered version ran. One hundred and five stable patients on clozapine, olanzapine or risperidone were randomised to CX516 900 mg three times daily or placebo for four weeks; ninety-five completed. The primary endpoint was change in a composite cognitive score. There was no difference from placebo on the composite, or on any individual cognitive test, at week 4 or at week 8. The between-group effect size in clozapine-treated patients was minus 0.19. The placebo group improved more than the drug group on the PANSS total score ^[11].

Nineteen patients gave large effect sizes; one hundred and five gave none. That is not a story about ampakines specifically, it is the ordinary behaviour of small studies, and it is why a pilot with a big effect size should be read as a reason to run a real trial rather than as a result.

The farampator study is the other one worth reading in full, because its finding is genuinely odd. Sixteen healthy elderly volunteers received 500 mg of farampator or placebo in a double-blind crossover. Short-term memory improved unequivocally. Episodic memory appeared to get worse. Side effects were headache, somnolence and nausea; the participants who had side effects had higher plasma drug levels, and those same participants performed significantly worse on memory than the ones who did not ^[12]. A drug with a dose-dependent side-effect burden that tracks worse cognition is a difficult therapeutic proposition regardless of what it does to short-term memory.

Alzheimer's disease was the largest test. LY451395 in 181 patients with mild to moderate disease produced no mean change from baseline on the ADAS-Cog at four weeks or eight weeks; the only measure that moved at all was the Neuropsychiatric Inventory total score ^[13]. Org 26576 in depression was explicitly exploratory, not a powered efficacy trial, and reported changes that were numerically but not statistically better than placebo ^[14].

So what went wrong? Nobody knows, and that is the honest answer. Three possibilities have never been separated. The molecules tested in humans were overwhelmingly type II, whose prolongation is kinetically capped by design, which is what made them safe enough to dose and may be what made them too weak to work ^[2]. Or the receptor-level effect may be real and simply not translate through the auxiliary subunits and network architecture of an intact human brain ^[3]. Or the underlying premise, that turning up existing excitatory transmission produces better cognition, may just be wrong. No adequately powered trial of a type I compound against a cognitive primary endpoint has been published, so the most obvious experiment in the class has not been done.

Sunifiram is the big one. Sunifiram is routinely marketed as a potent ampakine, and the best mechanistic work says the opposite. In mouse hippocampal slices it enhanced long-term potentiation in a bell-shaped dose-response curve peaking at 10 nanomolar, and that enhancement was blocked by 7-chlorokynurenic acid, an antagonist at the glycine site of the NMDA receptor, and not by ifenprodil. Pre-loading the slice with a high concentration of glycine abolished the effect entirely. AMPA receptor phosphorylation did rise, through CaMKII, but as a consequence of NMDA-site engagement rather than through any direct action on the AMPA receptor ^[16]. Calling it an ampakine gets the receptor wrong.

Unifiram and sapunifiram come from the same Florentine chemistry lineage and carry the same label for the same reason, which is that they sit next to the racetams on vendor lists. Neither has published human data of any kind. See research chemicals.

Noopept is neither a racetam nor an ampakine. It is a dipeptide ester, its main brain metabolite is the endogenous cyclic dipeptide cycloprolylglycine, and nothing in its published pharmacology involves allosteric modulation of AMPA receptors. It is filed here purely by association.

The general point is the one worth keeping. Ampakine is a mechanistic claim, not a category of vibe. It asserts that a molecule binds the dimer interface of an AMPA receptor and alters its gating kinetics, and that assertion is testable in a single patch-clamp experiment. When a compound is described as an ampakine, the reasonable question is which experiment showed it, at what concentration, and whether that concentration is one an oral dose could ever reach. For aniracetam the answer is a real experiment at a high concentration ^[7]; for sunifiram the answer points at a different receptor entirely ^[16]; for several others there is no answer at all. None of this is medical advice.