Glutamate receptors (NMDA, AMPA, kainate, mGluR)

Glutamate receptors divide cleanly into two camps. Ionotropic receptors (iGluRs) are the channels themselves: glutamate binds, the pore opens, cations cross within milliseconds. The three families are NMDA (GluN1, GluN2A to GluN2D, GluN3A and GluN3B), AMPA (GluA1 to GluA4) and kainate (GluK1 to GluK5). All are tetramers, assembled as a dimer of dimers rather than the pentameric ring used by nicotinic and GABA-A receptors, and all differ sharply in kinetics and in how much calcium they pass ^[1][2].

Metabotropic receptors (mGluRs) are family C GPCRs. Rather than conducting ions they bind glutamate inside a large extracellular clamshell and relay the signal through G proteins, changing excitability and synaptic strength over slower timescales. They work as obligate dimers, unusual among GPCRs and part of why their allosteric pharmacology is so productive ^[18][19]. Short version: iGluRs carry the message, mGluRs set the volume.

Clearance is where glutamate stops resembling any other transmitter, and a lot of pathology lives here. There is no degrading enzyme in the cleft; nothing plays the role acetylcholinesterase plays for acetylcholine. Glutamate is removed entirely by transporters, mostly by astrocytes through EAAT2, which convert it to glutamine, ship it back to the neuron and let the neuron convert it back. That glutamate/glutamine cycle is the whole recycling loop ^[3].

The numbers explain the stakes. Extracellular glutamate is held around one to three micromolar while the concentration inside a nerve terminal is roughly ten thousand times higher. A gradient that steep means uptake failure is a collapse rather than a drift, which is why ischaemia, in which transporters run backwards as the energy supply fails, floods the extracellular space within minutes ^[3][36]. It follows that there is no useful raise your glutamate strategy: the system is limited by receptors, by clearance and by where the receptors sit, never by supply.

Each row below is one receptor class. The ionotropic types differ by speed and by calcium handling; the metabotropic types are grouped by G protein coupling and by where on the synapse they sit ^[1][19].

Read the defining feature column as what makes each class drugable. NMDA has three places to intervene, the channel, the glutamate site and the co-agonist site, which is why its pharmacology is so crowded. AMPA has nothing inside the pore worth targeting, so its drugs are allosteric. mGluRs have an allosteric pocket that differs between subtypes, the only reason subtype-selective glutamate drugs exist at all ^[18].

An NMDA receptor is an obligate heterotetramer: two GluN1 subunits and usually two GluN2 subunits. The division of labour is worth memorising. GluN1 binds glycine or D-serine, not glutamate; GluN2 binds glutamate. Both sites must be occupied, and the cell must already be depolarised enough to expel magnesium, before the channel conducts. Three conditions, one channel ^[1][2].

The magnesium block was reported independently by two groups in the same year, which is a good sign a result is real ^[4][5]. The co-agonist requirement came three years later, when glycine was shown to be necessary rather than merely helpful ^[6], and the identity of the endogenous co-agonist took longer still: D-serine, made by astrocytes from L-serine, fills that site at many forebrain synapses ^[7]. That is why D-Serine, Glycine, Sarcosine (a glycine transporter inhibitor) and Bitopertin are studied as ways to raise NMDA function without touching the glutamate site at all.

Which GluN2 subunit is present changes almost everything measurable: how long the channel stays open, how strongly magnesium blocks it, how much calcium enters, and which scaffolding proteins it recruits. Forebrain synapses also undergo a developmental switch from GluN2B towards GluN2A that shortens the response and is thought to close critical periods ^[8][9].

Location matters as much as composition, and this is the least intuitive fact on the page. Activating NMDA receptors inside the synapse drives CREB and pro-survival signalling; activating the ones outside it drives CREB shut-off and cell death programmes. Same protein, opposite outcome, decided by where it sits ^[10]. That is the best current account of why NMDA blockade is neuroprotective and cognitively damaging at once, and why usefulness here depends on which receptors a drug reaches rather than how many.

It also explains the channel blockers, which all bind the same site inside the open pore and behave completely differently. The axis is dwell time. Memantine and Amantadine are low-affinity, fast off-rate, strongly voltage-dependent blockers; they leave briskly enough that a normal synaptic burst clears them, so ordinary transmission survives while chronically activated receptors stay blocked ^[34]. Ketamine has higher affinity and dwells longer, which is where dissociation comes from; MK-801 sits at the extreme and is a laboratory tool rather than a medicine. One kinetic parameter separates an approved Alzheimer's drug from a research toxin.

A second route to selectivity targets one GluN2 subunit rather than the pore. Traxoprodil, also known as CP-101,606, is GluN2B-selective, and in a small randomised trial in treatment-refractory depression it produced an antidepressant response without a dissociative reaction ^[38]. That is the cleanest evidence available that the antidepressant effect and the dissociation are separable, and it is the reasoning behind Rislenemdaz and the rest of the subunit-selective line.

AMPA receptors carry nearly all fast excitatory transmission, and their most important property is decided by a single edited letter. The GluA2 subunit is transcribed with a glutamine at a key pore position, and an RNA editing enzyme changes it to an arginine before the protein is made. Any receptor containing edited GluA2 is calcium impermeable; any receptor lacking it is calcium permeable ^[15][16]. Editing runs at essentially 100% in healthy brain, so most AMPA receptors pass sodium and not calcium, and a fall in editing efficiency has been linked to motor neuron disease and to ischaemic injury.

Two more layers make the same four genes behave like many receptors. Flip and flop are alternative splice variants of one exon that change desensitisation kinetics, so a single gene yields a fast receptor and a slow one depending on the cell ^[14]. And AMPA receptors do not travel alone: auxiliary subunits, chiefly the TARP family, control trafficking, gating and even pharmacology, so behaviour is not a property of the pore-forming subunits by themselves ^[2][13].

That trafficking is not a detail; it is the mechanism of learning. Strengthening a synapse during LTP consists largely of inserting more AMPA receptors, and weakening it consists of removing them ^[13]. NMDA receptors decide when it happens; AMPA receptors are what changes.

Ampakines are positive allosteric modulators of that receptor. The first clean demonstration used Aniracetam, which increased the conductance change at quisqualate and AMPA receptors without altering agonist affinity or the channel's ion selectivity, and left kainate, NMDA and GABA responses alone ^[33]. The detail usually left out is that the effect appeared above 0.1 mM, a high concentration.

Later ampakines split along a line that matters: high-impact modulators slow deactivation aggressively and carried a seizure liability, while low-impact ones only nudge it. TAK-653 is the careful version, potentiating the receptor only when glutamate is already present and carrying minimal agonist activity of its own ^[31]. In healthy volunteers it produced dose-related central effects at tolerated doses, so the pharmacology does reach the human brain ^[32]. Older members here include CX-516, CX-614, Farampator, Org 26576 and Tulrampator.

Kainate receptors are the least tidy of the three. They are slower, sit both pre- and postsynaptically, and regulate transmitter release rather than carrying the main signal. They also do something no other ionotropic glutamate receptor does: part of their effect runs through G proteins, so a receptor classified as a channel behaves partly like a metabotropic one ^[17]. Topiramate blocks them among several other actions, which is a fair illustration of how little selective pharmacology this class has.

The reason glutamate receptors matter so much for cognition is long-term potentiation (LTP), a lasting strengthening of synapses that is widely treated as the cellular basis of learning. At the classic hippocampal synapse it depends on the NMDA receptor acting as a coincidence detector: AMPA receptors depolarise the postsynaptic cell first, which expels the magnesium plug, and only then does calcium enter and set off the biochemical changes that lock in a stronger synapse ^[11].

That calcium signal is a switch rather than a current. It activates CaMKII and other kinases, drives more AMPA receptors into the membrane, and enlarges the dendritic spine itself. The history is worth knowing: the field spent roughly twenty years arguing whether the change was presynaptic or postsynaptic before settling largely on postsynaptic AMPA receptor trafficking ^[12][13]. It is a good example of a mechanism that felt settled long before it was.

The same machinery feeds growth signalling. Rapid antidepressant effects in animals depend on a burst of glutamate reaching AMPA receptors, which drives BDNF release and mTORC1-dependent synapse formation within hours, a chain that fails if AMPA receptors are blocked ^[28][30]; see BDNF and neuroplasticity.

Metabotropic receptors shape all of this from the sidelines, and group I mGluRs can trigger a competing long-term depression that weakens a synapse instead ^[19][20]. Plasticity is therefore not one process with a volume knob; it is at least two opposed processes, and which one a pattern of activity produces depends on the receptors present and on how much calcium arrives.

Eight metabotropic receptors sort into three groups by sequence, coupling and pharmacology, and the grouping has held up since it was laid out ^[18][19]. The position of group II is the elegant part. Because mGluR2 and mGluR3 sit outside the active zone, ordinary transmission does not reach them; they respond only when release has been heavy enough to spill glutamate beyond the synapse. That makes an agonist there a selective brake on excessive release rather than a general suppressant, which is why the group looked so promising in psychosis and anxiety ^[19][24].

Two large clinical programmes tested the group I and group II ideas properly. Both failed, in ways the preclinical work did not predict, and both are worth studying for that reason.

Fragile X and mGluR5. The mGluR theory held that without the FMRP protein, group I signalling runs unopposed, so blocking mGluR5 should correct the consequences ^[20]. It was an unusually clean hypothesis with strong animal support. Two randomised, double-blind, placebo-controlled phase 2b trials of Mavoglurant, in 175 adults and 139 adolescents, both missed the primary behavioural endpoint, and the authors stated plainly that under the conditions tested they could not confirm the theory ^[21]. Later treatment, longer placebo run-ins and younger patients were proposed afterwards; none has been demonstrated.

Schizophrenia and mGluR2/3. Pomaglumetad methionil, a prodrug of a group II agonist, separated from placebo in a phase 2 trial and was described at the time as the first non-dopaminergic antipsychotic mechanism to work ^[22]. In a 678-patient phase 3 comparison it lost to aripiprazole on symptom change while producing less weight gain, and the programme did not continue ^[23]. The glutamate hypothesis of schizophrenia is not thereby refuted; it remains the best account of why NMDA antagonists reproduce the illness so faithfully ^[24][25]. What failed was one specific way of acting on it.

The pattern is the lesson. Group I negative modulators such as MTEP, Fenobam and Basimglurant have excellent target engagement and repeatedly thin clinical results. In this family the preclinical rationale has outrun the clinical outcome more than once, and any new mGluR claim should be read with that history attached.

Memory. Because NMDA-dependent LTP is the leading model of learning, drugs touching this system are perpetually studied for cognition, and glutamate-tuning agents such as Dimiracetam belong to that effort ^[11][12]. No compound acting here has an established pro-cognitive effect in healthy people; the strongest human evidence in this family is for treating disease rather than improving normal function.

Excitotoxicity. Too much glutamate lets too much calcium through NMDA channels, and sustained calcium entry activates proteases, nitric oxide synthase and mitochondrial death pathways; a well-characterised mechanism in stroke, trauma and neurodegeneration ^[36]. What followed is a cautionary tale. NMDA antagonists protected neurons convincingly in animals and then failed in every major human stroke and head injury trial, because the doses needed to protect also caused psychotomimetic effects and hypotension, and because the treatment window closed before patients reached hospital ^[37]. The biology was right and the drug strategy was wrong, a distinction worth carrying to any neuroprotection claim.

Memantine is the exception, and it works because of how it blocks rather than that it blocks. Low affinity and a fast off-rate let it leave the channel quickly enough for ordinary synaptic bursts to punch through, while chronically activated receptors stay blocked ^[34]. It is approved for moderate to severe Alzheimer's disease on modest but reproducible benefit ^[35], and Nitromemantine tries to sharpen the same selectivity further.

Rapid antidepressants are the biggest thing to come out of this receptor in fifty years. A small crossover study in 2000 found a single sub-anaesthetic ketamine infusion improved depression within hours ^[26], and the 2006 randomised trial in treatment-resistant depression made the result impossible to ignore ^[27]. The proposed mechanism runs through a burst of AMPA signalling, BDNF release and mTORC1-dependent synaptogenesis rather than through NMDA blockade itself ^[28][30], which is exactly why AMPA potentiators are pursued as better-tolerated successors ^[31]. Esketamine is approved; Arketamine, Esmethadone, Apimostinel and Zelquistinel are attempts to keep the effect and drop the dissociation.

Schizophrenia. The glutamate hypothesis was born from one human observation: sub-anaesthetic ketamine in healthy volunteers reproduces positive symptoms, negative symptoms and the cognitive profile of the illness at once, which no dopaminergic drug does ^[25]. Two decades of work turned that into a detailed model of NMDA hypofunction on inhibitory interneurons producing downstream glutamate disinhibition ^[24]. The model is strong; the drugs built from it, so far, are not.

Whether NMDA blockade is required for ketamine's antidepressant effect at all. One line of work reports that the metabolite (2R,6R)-hydroxynorketamine produces antidepressant-like effects in mice without inhibiting NMDA receptors ^[29], the effect depends on AMPA receptors rather than on the depth of NMDA blockade ^[30], and Lanicemine, a low-trapping NMDA blocker, did not replicate the clinical benefit. Those three facts do not fit a simple receptor-occupancy story, and no account currently reconciles them.

Whether an AMPA potentiator translates. TAK-653 has shown central pharmacodynamic effects in healthy volunteers at tolerated doses ^[32], which is more than most ampakines achieved, and still has no positive controlled efficacy result in a psychiatric indication ^[31]. The class has produced measurable brain effects and no approved drug across decades.

Whether aniracetam's AMPA mechanism operates in a person. The in vitro result is clean and was obtained above 0.1 mM, and most of an oral dose reaches the blood as a metabolite rather than as the parent ^[33]. Nobody has shown the mechanism does work at human exposures.

Why the mGluR programmes failed. Wrong age at treatment, wrong endpoint, wrong species or a wrong theory are all live explanations for fragile X ^[20][21], and the same list applies to group II agonism in schizophrenia ^[22][23]. Choosing between them requires trials nobody has funded.

Whether location or subunit composition is the real variable. GluN2B is enriched outside the synapse, so synaptic versus extrasynaptic and GluN2A versus GluN2B predict the same data in most experiments, and drug design depends on which one is doing the work ^[9][10]. In the same corner, GluN3 subunits lower current and calcium flux when included and are developmentally regulated, and their normal physiological role remains poorly characterised ^[2].

Whether excitotoxicity is treatable in humans at all. The mechanism is well established in tissue ^[36] and every attempt to block it clinically has failed for reasons about pharmacology and timing rather than about the mechanism ^[37]. Memantine is the only partial answer anyone has ^[34].