GABA and Neural Inhibition: The Brake, and What Benzodiazepines Do to Memory

Introduction

You are wheeled into a room. Someone puts a cannula in the back of your hand and pushes a small volume of clear liquid. A nurse asks whether you are comfortable. You say yes. You answer two more questions. You watch the screen. Twenty minutes later somebody helps you into a chair, tells you the results, gives you a leaflet, and walks you to the door where a friend is waiting.

Then you wake up in the car with no idea how you got there.

The strange part is not that you fell asleep, because you did not. You were awake for all of it.

There is video of some of these procedures. Patients talk, follow instructions, laugh at jokes, complain about the cold. Afterwards, nothing. Not a fuzzy memory, not a partial one. A clean absence where an hour should be.

Most explanations you will find say the drug caused memory loss. That phrase quietly assumes something that is almost certainly false.

Loss implies a thing you had and then misplaced. What actually happened is stranger and, once you see it, much simpler. The hour was never written down. Your hippocampus was online, your eyes were open, your mouth was working, and the part of the machinery that turns experience into a record had been switched off.

The molecule at the centre of this is gamma-aminobutyric acid, which everybody calls GABA. It is the main inhibitory neurotransmitter in your brain [1]. It is what your brain uses to say no.

This article is about what that brake is, how it works at the level of a single channel in a single membrane, and what happens when a drug leans on it.

Along the way it runs into a mouse that got smarter when a piece of its brake was deleted, a brake that works backwards in newborns, a group of patients who remembered things that never happened, and a question about long-term harm that the field genuinely has not settled and pretends to less often than it should.

The Brake, and What It Actually Does

Start with the obvious problem. Neurons excite each other. One fires, releases glutamate, and makes the next one more likely to fire. If that were the whole system, the first neuron to fire anywhere would set off every neuron connected to it, and then every neuron connected to those. A single thought would end in a seizure. The brain would have exactly one behaviour and you would not survive it.

So there has to be a no. Inhibition is not a safety feature bolted onto an excitatory brain. It is half the design.

The cells that deliver it are interneurons, and they are not one thing. They are dozens of distinct cell types with different shapes, different firing patterns, and crucially different targets on the cells they inhibit [2]. Some wrap the cell body of their target and can veto its output entirely. Others touch only the far tips of its dendrites and shave off particular inputs while leaving the rest alone. Others inhibit other inhibitory cells, which means their net effect is to release the brake.

They are a minority of the neurons in your cortex. They are nothing like a minority of the control [3].

It helps to stop thinking of them as brakes on individual cells. A cell that inhibits the body of a neuron is a veto. A cell that inhibits one branch of its dendrites is an editor, removing one line of an incoming argument and leaving the rest standing. Those are completely different operations, and calling both of them inhibition hides how much is going on.

One clarification, because a lot of writing on this topic flattens it. GABA runs inhibition in your brain. Further down, in the spinal cord and brainstem, the main inhibitory transmitter is glycine [4]. Same job, different molecule, different receptors. When you read that GABA is the inhibitory transmitter of the nervous system, that is a compression of something more interesting.

If the way one neuron talks to another is not already familiar, our piece on how neurons communicate across a synapse is the ground floor for everything below.

A Channel That Lets Chloride In

Here is the actual mechanism, and it is worth slowing down for, because everything else in this article is a consequence of it.

A neuron sits at a negative voltage inside relative to outside. To fire, it has to become less negative, and cross a threshold. Excitation pushes it toward that threshold. Inhibition pushes it away.

GABA pushes it away by opening a hole for chloride.

The GABA-A receptor is five protein subunits arranged in a ring around a central pore [5]. When two GABA molecules bind, the ring twists and the pore opens, and it is selective for chloride ions. Chloride is negatively charged and there is more of it outside the cell than inside, so it flows in. The inside of the neuron gets more negative. The threshold gets further away. A message that would have made the cell fire now does not.

That is the whole trick. Not a signal that says stop. A hole that makes firing harder.

Structural biology has now seen this directly. Cryo-electron microscopy has resolved human GABA-A receptors sitting in a lipid membrane rather than dissolved in detergent [6], and more recently the actual receptor assemblies pulled out of brain tissue rather than the ones we build in cell lines [7]. That distinction matters more than it sounds. For years the field studied combinations that were convenient to express, and it was a fair question how many of them the brain actually builds.

There is a second family, the GABA-B receptors, which are not channels at all. They are G-protein coupled receptors, which means GABA binding sets off a slower chain of chemistry inside the cell that eventually changes potassium and calcium channels. Slower, longer lasting, and pharmacologically a different world.

Look at the drug column. Nearly every sedative, hypnotic and anticonvulsant humanity has stumbled onto converges on the top row.

Alcohol is in there. So is the general anaesthetic that put you under for surgery, which is the subject of our article on what anaesthesia does to memory. That convergence is not a coincidence. If you want to turn a brain down, this is the volume knob it has.

1977: The Year Somebody Found the Valium Receptor

Benzodiazepines came before the explanation. Chlordiazepoxide and diazepam were in clinics and in millions of medicine cabinets while nobody could say what they did.

Then in 1977 a specific binding site for benzodiazepines was demonstrated in the central nervous system [8]. The drug was not doing something diffuse to membranes. It was binding a particular target, in a particular place, and that target turned out to be part of the GABA-A receptor.

Here is the detail that almost nobody outside pharmacology knows, and it changes how you should think about these drugs entirely.

The benzodiazepine site is not the GABA site.

They are physically separate. GABA binds at one interface between subunits; the benzodiazepine binds at a different interface on the same receptor [9]. And a benzodiazepine, on its own, cannot open the channel. It has no ability to inhibit anything. What it does is make the receptor more responsive to the GABA your own brain is already releasing.

The technical term is positive allosteric modulator. The plain English version is better: it does not add a signal, it turns up the gain on a signal you were already sending.

Why this matters practically. A drug that opened chloride channels by itself would flatten everything, everywhere, including your breathing.

A drug that only amplifies existing GABA release follows the brain's own pattern of where inhibition is already happening. That is why benzodiazepines are, in isolation, remarkably difficult to die from, and also why they become dangerous when combined with opioids, which suppress breathing through an entirely different route.

For two decades after 1977 there was a puzzle sitting on top of this. Benzodiazepines do at least five separable things. They reduce anxiety. They cause sedation. They relax muscles. They stop seizures. And they interfere with memory. If they all worked through one receptor, why did they not always arrive together?

The answer came from mutant mice in 1999 [10]. Researchers made point mutations that rendered specific subunits insensitive to diazepam while leaving them otherwise functional, then asked which effects survived. The effects came apart. Sedation tracked one subunit. Anxiety reduction tracked another.

It is worth pausing on how unusual that experiment was. Instead of building a more selective drug and seeing what it did, they rebuilt the receptor so that an existing drug could no longer touch one part of it, then asked which of the drug's effects went missing.

The animal is the instrument. That design is why the conclusion held.

That result reframed everything. There is no such thing as the benzodiazepine effect. There are several effects, riding on different receptor subtypes, which happen to be bundled together in the drugs we have [11]. And where a subunit sits in the brain determines what happens when you modulate it, which is why mapping the distribution of the subunits [12] across regions and cell types [13] mattered as much as it did.

Hold onto that. It is the key to the memory question, and we will come back to it.

Two Kinds of Quiet

Through the 1980s the picture of inhibition was: a synapse fires, GABA crosses, a brief current flows, done. Fast, discrete, event-shaped.

Then people started recording cells more carefully and found something else underneath. A standing chloride conductance that was there all the time, not tied to any particular synaptic event [14]. It was small, it was constant, and it had been treated as baseline noise.

Careful recording is what turned it up, and there is a small lesson in that. Baseline is a decision, not an observation. When you subtract a steady current as background, you have already decided it is not part of the signal.

It was not noise. It was a second kind of inhibition, and the framework that organised it has been the field's reference point ever since [15].

Phasic inhibition is the event. A synapse releases a puff of GABA, it hits receptors packed directly opposite, they open, a brief current flows, and the transmitter is cleared. It blocks a specific message at a specific moment.

Tonic inhibition is the background. Some GABA-A receptors sit outside the synapse entirely, and those extrasynaptic receptors are exposed to whatever low concentration of GABA is drifting around in the extracellular space [16]. They are more sensitive and they do not desensitise as quickly, so they hold a persistent, low-level brake on the cell [17].

The distinction is not academic bookkeeping. Think about what a persistent background brake actually is. It is a gain control. It sets how excitable a whole population of cells is before any particular message arrives.

Which means changing it changes what the circuit is capable of, not just what it is doing right now. Turn tonic inhibition up in the dentate gyrus of a mouse and you impair long-term potentiation and memory [18], the strengthening process described in our article on how long-term potentiation works. Turn excessive tonic inhibition down after a stroke and motor recovery improves [19]. Same knob, opposite directions, depending on where the set point started.

You can feel the difference between the two if you think about what each one would sound like. Phasic inhibition is somebody interrupting one sentence. Tonic inhibition is the volume of the room.

And neurons are not the only ones with a hand on it. Astrocytes release GABA too, and they contribute to the ambient pool that drives tonic inhibition [20]. The tonic level also moves with the ovarian cycle in mice, shifting seizure susceptibility and anxiety-related behaviour along with it [21]. Inhibition is not a constant. It is a set point that drifts.

Think about what that implies for anything you read about GABA levels. A single number for how much GABA a brain contains tells you very little, because the same molecule is doing two jobs on two timescales through two populations of receptors, and a drug or a disease can move one without touching the other.

Other things bind this receptor and act mainly on the tonic side. Neurosteroids, made in the brain from cholesterol, selectively enhance tonic inhibition through delta-subunit receptors [22], a property recognised as far back as the mid-1980s when steroid metabolites turned out to modulate the receptor much like barbiturates do [23]. Those compounds have since become drugs in their own right [24].

Remember the subunit lesson from 1999. Now notice which subunits show up in the tonic column. One of them is alpha-5.

The Brake That Used to Be an Accelerator

Before the memory story, one detour, because it is the cleanest illustration of how contingent all of this is.

Everything above depends on one assumption: that chloride flows into the neuron when the channel opens. That is not a property of the receptor. It is a property of the chloride gradient across the membrane, and gradients can point either way.

In the immature brain, they point the other way.

Young neurons have high internal chloride, because they express a transporter called NKCC1 that pumps chloride in [25]. Open a chloride channel in a cell like that and chloride flows out. The cell becomes less negative, not more. GABA depolarises it.

In the developing brain, in other words, the brain's main inhibitory transmitter is excitatory [26].

Take a second with that, because it sounds like a typo and is not. The same transmitter, binding the same receptor, opening the same channel, produces the opposite result. Nothing about the receptor has changed. The direction of the current has.

The switch happens when a second transporter, KCC2, comes online and starts pumping chloride out, dropping internal chloride below external and flipping the direction of flow [27]. Delete KCC2 and even early synaptic inhibition fails [28]. The whole developmental sequence turns on the balance between these two transporters [29].

This is not a piece of trivia about embryos. It has three consequences that reach into adult medicine.

It also reframes what inhibition is for early on. A depolarising GABA signal in an immature circuit is not a failed brake. It is a growth signal, helping young neurons wire themselves up before there is anything to inhibit.

First, it explains why drugs that quiet an adult brain do not behave the same way in a newborn one, and why NKCC1 activity facilitates seizures in the developing brain [30].

Second, the switch can run backwards. In tissue resected from people with temporal lobe epilepsy, chloride homeostasis is measurably disturbed, and GABA's effect in that tissue is not what it should be [31]. A circuit that has lost its chloride gradient has lost its brake, whatever the receptors are doing.

That second point is worth dwelling on, because it inverts how people usually think about drug resistance. A patient whose seizures stop responding to a drug that boosts GABA is not necessarily short of GABA, and the drug is not necessarily failing to bind. The receptor may be working exactly as designed, opening a channel onto a gradient that no longer points the right way.

Third, the same reversal shows up after spinal cord injury, where KCC2 downregulation contributes to spasticity [32]. Muscles that will not relax, traced back to a transporter.

The brake is not built into the receptor. It is built into a gradient that something has to maintain.

The Mice That Got Smarter

Now the memory question, and it starts somewhere that sounds like a mistake.

In 2002 a group produced mice lacking the alpha-5 subunit of the GABA-A receptor. Alpha-5 is unusual in that it is concentrated heavily in the hippocampus, the structure that binds experience into episodes. If you delete a piece of the inhibitory machinery from the memory structure, the obvious prediction is that memory gets worse.

It got better. The knockout mice outperformed normal mice on hippocampus-dependent learning, with measurably altered GABAergic transmission in area CA1 [33].

The result was not what anybody was looking for, which is usually a sign that it is telling you something.

Read that again, because it is doing more work than it appears to.

It means a normal mouse is carrying a brake on hippocampal learning that it does not strictly need for that task. Inhibition in the hippocampus is not just noise suppression. It costs something, and evolution has apparently decided the cost is worth paying.

A second group came at it from a different angle the same year, using point-mutated mice and trace fear conditioning, a task that requires the hippocampus to bridge a gap in time. Alpha-5 receptors were specifically involved [34]. Two labs, two methods, one subunit.

Then the pharmacology caught up. If deleting alpha-5 improves learning, a drug that selectively turns alpha-5 receptors down should do the same thing without genetic engineering. It does. An inverse agonist selective for alpha-5 containing receptors enhanced cognition in rats [35].

And then the experiment that matters most for this article.

Researchers gave the same class of compound to rats in a Morris water maze and varied when they gave it. Before the first trial. Before the second trial. Immediately after the first trial. The improvement appeared when the drug was given before a trial. It did not appear when the drug was given immediately after [36].

That timing is a fingerprint. Give a drug before learning and you affect encoding. Give it after learning and you affect consolidation, the process that stabilises what was just encoded. This compound moved encoding and recall. It did not move consolidation.

Hold that shape in mind. We are about to see its exact mirror image in humans, produced by a drug pushing in the opposite pharmacological direction.

A caution before we do. Alpha-5 selective compounds have been through preclinical and early clinical work, and the honest summary is that the preclinical results are strong and consistent and no cognitive enhancer has come out the other end [37]. The approach remains active, including work rescuing behavioural and electrophysiological abnormalities in a mouse model of a genetic syndrome [38], and the broader role of these receptors in hippocampus-dependent cognition is still being mapped [39]. Nobody should read this section as a memory pill being around the corner. It is not.

What it is, is proof that the memory effect of these drugs is not a vague consequence of drowsiness. It has an address.

The Hour You Were Awake For

Back to the room, the cannula and the missing hour.

Anterograde amnesia after a benzodiazepine is not a rare reaction or an overdose phenomenon. It is a property of the class, with onset and duration varying by which drug, what dose, and which route [40]. Anterograde means forward in time: the gap starts when the drug takes effect and covers what comes after.

Which immediately raises the question this article is really about. Why forward?

If the drug were dissolving memories, the gap would sit on both sides of the injection. Things you learned an hour before would be at risk too. That is what retrograde amnesia looks like, and it is what happens after a serious head injury.

That is not the pattern. The pattern is that material presented before the drug survives, and material presented after the drug does not [40]. Which points at encoding, not storage, and not retrieval.

You were awake. You were talking. Nothing was being written down.

The wording matters here in a way that is not pedantry. If somebody tells you a drug wiped their memory of a procedure, the word wiped is doing damage. It suggests there is a file somewhere in a damaged state, and that with the right cue it might come back. It will not, because there is no file.

Now put that beside the rat result from the previous section. A drug that turns alpha-5 receptors down improves encoding and recall and leaves consolidation alone. A drug that turns GABA-A receptors up destroys encoding and leaves already-stored material alone. Two literatures, two species, opposite pharmacological directions, converging on the same stage of memory.

That convergence is the strongest thing this article has to offer, and it is why the framing matters. Where the failure happens is not a detail. It is the difference between something being taken from you and something never having existed.

If you want the machinery on the other side of that arrow, our article on how the hippocampus decides what to remember covers what is supposed to be happening in that window.

The detail that unsettles people. A study of 100 women undergoing elective caesarean delivery under spinal anaesthesia tested this directly. Six picture cards were shown at one-minute intervals immediately after birth, and then midazolam was given or not. Retrograde recall of the cards was lower in the midazolam group, 77.0 percent against 87.7 percent, which suggests a brief retrograde window rather than none at all [41].

Two things have to be said about that study in the same breath as its result. It was not randomised. Which group a patient ended up in followed her own preference, and preferences are not random. So treat the size of that difference with care.

The second finding is the one that stays with you. The midazolam group produced more spurious reports of events during the procedure that had not happened.

The gap does not stay a gap. Something fills it.

Ask somebody what happened during a window their brain never recorded and they will not usually say nothing, because that is not how remembering works from the inside. Our article on how memories get rewritten each time they are recalled deals with the same uncomfortable property from a different direction.

It is also worth being clear about what this does not mean. Nothing here suggests that a person under sedation is not experiencing what is happening to them. They are awake. They feel discomfort, they hear what is said, they respond. The recording is what fails, not the experience.

This is not new and it is not confined to hospitals. In 1987 a report in a major medical journal described travellers who took a short-acting benzodiazepine to sleep on a flight and arrived with hours missing, walking through airports and conversations they could not later account for [42]. The phenomenon got a name that stuck. High doses of the same drug were reported to produce dense anterograde amnesia [43], and the amnestic effect is deliberately used in medicine, where a drug given before a procedure is meant to reduce anxiety and leave no memory of it [44].

That last point deserves emphasis, because it is easy to read all of this as harm. In a procedure room the amnesia is often the goal. A patient who is calm during an unpleasant procedure and carries no memory of it afterwards has had a better experience than one who remembers every minute. That is why these drugs are still used the way they are, and why anaesthesia research continues to work on getting the balance right [45], including the question of what excess inhibition costs in the hours and days afterwards [46].

So whether the amnesia is a side effect or the point depends entirely on who is holding the syringe and why. That is unusual. Most drug effects are one or the other.

One more piece of evidence separates the two effects that keep getting merged. Benzodiazepines slow the peak velocity of your eye movements, and that slowing has been meta-analysed across 30 studies and 45 effect sizes [47]. Saccadic slowing is a clean sedation readout. It can be measured independently of anything to do with memory. Sedation and amnesia are not the same axis, they can be quantified separately, and the alpha-5 evidence says the memory axis has its own mechanism. Sedation contributes. It is not the explanation. And the honest position, stated by the people who review this literature, is that the cellular and molecular mechanisms of benzodiazepine-induced amnesia are not fully understood [40].

There is one more reason the drug-state matters. What you can recall depends partly on whether your internal state at retrieval matches your state at encoding, an effect covered in our article on state-dependent memory. That effect is real, and it is not what is happening here. State-dependent forgetting means the trace exists and you cannot reach it. This is the other thing.

Sedation Is Not Sleep

This distinction gets lost constantly, and it is the one that matters most for anyone taking something to sleep.

A benzodiazepine will put you to sleep. That does not mean it gives you sleep.

Sleep is not a single off state. It is a structured sequence of stages, and the deep slow-wave portion of it is heavily implicated in stabilising what you learned during the day, which our article on how sleep consolidates learning goes into properly.

Drugs acting at this receptor change that structure. The cleanest demonstration is an old one and it runs backwards: give the benzodiazepine antagonist flumazenil, which blocks the benzodiazepine site without doing anything itself, and slow-wave sleep changes [48]. If blocking the site alters sleep architecture, the site is part of how sleep architecture is set.

What that means for memory has been tested directly, though in small samples. Twenty-two healthy participants went through three conditions in a counterbalanced crossover design, comparing two benzodiazepine-receptor-agonist hypnotics against sleep-dependent memory consolidation [49]. Twenty-two people is not many. Treat it as a signal, not a settled quantity.

Twenty-two people is also a reminder of how this field is built. Much of what we know about drugs, sleep and memory in humans comes from small crossover studies in healthy volunteers, because the alternative is dosing large numbers of people with a controlled substance and following them for years.

Small samples are not sloppiness. They are the shape of what is ethically and practically possible, and the right response is to hold the conclusions loosely rather than to ignore them.

The picture is not one-directional, and an article that told you these drugs simply wreck sleep would be misleading you. A 2026 study in a mouse model reported that zolpidem restored sleep and reduced amyloid [50]. That is a mouse, in a model of a disease, and it should be read as a mouse in a model of a disease. But it is a real result pointing the other way, and pretending it does not exist would be exactly the kind of tidy story this field does not support.

There is also what happens to the receptor itself over weeks. Prolonged exposure to diazepam reduces the number of alpha-1 containing receptors on the cell surface and uncouples the benzodiazepine site from the GABA site [51]. That is tolerance with a mechanism attached. The cell is not being stubborn. It is adjusting a gain that something outside keeps pushing on, and it is why these drugs are generally prescribed for short periods [52].

Any change to a prescription is a conversation with the person who wrote it, not a conclusion to draw from an article. Stopping a benzodiazepine abruptly can cause seizures. That is the one piece of practical information in this piece, and it is the reason there is no other.

A Fifty-Year Argument, In Order

The pieces above did not arrive in a sensible order. Here is the sequence they actually came in.

Two things stand out in that list. The mechanism was found before it was understood, which is normal in pharmacology and almost never how it is taught. And the memory-specific result, the alpha-5 work, arrived twenty-five years after the receptor and is still not in most explanations written for patients.

The structural work is worth a closing note. It took until 2019 to see a human GABA-A receptor properly in a membrane [6], and until 2023 to see the combinations the brain actually assembles rather than the ones that grow well in a dish [7]. Meanwhile drugs targeting this receptor have been prescribed by the hundreds of millions of units per year since the 1960s. The gap between using something and seeing it is not a scandal. It is just how much of medicine has worked.

What Long-Term Use Does, and What Nobody Can Prove Yet

This is the part of the topic where the internet is least reliable in both directions, and it deserves to be handled carefully.

The question is whether long-term benzodiazepine use raises the risk of dementia. It has been studied repeatedly for two decades, and the results genuinely disagree.

Read the column on the right. Several datasets find an association [53] [54] [55] [56], including a 2025 case-control study reporting a stronger signal for drugs with long half-lives than medium ones [57]. One large community cohort found a link to mild cognitive impairment and not to dementia [58]. A population study with repeated brain imaging examined long-term risk alongside markers of neurodegeneration rather than asserting a cause [59].

And then there is the paper that makes the whole thing harder.

Using 40,770 dementia cases matched against 283,933 controls, a 2019 analysis showed how much of the reported association can be produced by the design of the studies themselves, specifically by how prevalent users are handled and when covariates are measured [60]. Not by the drug. By the shape of the analysis.

The confound has a name, and once you see it you cannot unsee it.

Early dementia does not announce itself as memory failure. It announces itself as insomnia, anxiety, restlessness and agitation, often years before anybody uses the word dementia.

Those symptoms get treated. What frequently gets prescribed for them is a benzodiazepine. So the prescription arrives before the diagnosis, faithfully, in the data, for a reason that has nothing to do with the drug causing anything.

This is called reverse causality, or protopathic bias. It does not prove the drug is harmless. It proves that the observational data cannot currently distinguish the two stories, which is a different and more honest claim.

None of this means the question is unanswerable. It means answering it properly needs study designs that observational prescription data cannot easily provide, and those are slow, expensive and hard to run. In the meantime, the gap between what the data shows and what gets written about it stays wide.

So the accurate summary is uncomfortable and short. The association is real in several datasets. It is absent or attenuated in others. A large methodological analysis shows the design alone can generate it. Anyone telling you benzodiazepines cause dementia is going beyond the evidence, and so is anyone telling you the link has been ruled out.

Some effects of long-term use are not contested at all, and it is worth separating them out. Driving performance and the cognitive skills underlying it are measurably affected in long-term users [61], a finding consistent enough to support formal evidence synthesis [62]. Benzodiazepine exposure in older adults after hospitalisation has been linked to mortality at six months [63]. These are not the dementia question and should not be folded into it.

Reducing long-term prescribing is an active field of clinical research, including trials of how to do it in ordinary primary care [64] and work on principles for doing it safely [65]. That work exists. It is for prescribers and their patients together. It is not instructions for a reader, and the risk profile of these medicines is a live policy discussion in its own right [66].

Older brains are a distinct case throughout, and our article on how memory changes with age covers the baseline these studies are measured against.

When the Balance Breaks

Everything so far has been about what happens when something interferes with inhibition. The more interesting question is why the brain needs so much of it in the first place, because the reason inhibition matters is bigger than pharmacology.

Excitation and inhibition are not two forces that happen to coexist. Balance between them is a computational requirement. Network models showed decades ago that when excitation and inhibition are matched, the network settles into an irregular, responsive regime that neither runaway excitation nor total suppression can produce [67]. The balance is what makes the system able to compute at all.

Inhibitory cells also generate rhythm. The gamma oscillation, the fast rhythm associated with attention and binding, is produced by networks of inhibitory interneurons [68]. Not by excitatory cells with inhibition trimming them. By the interneurons themselves, whose mutual inhibition sets the beat.

Which is why so many disorders show up here.

Epilepsy is the most direct case. It is what a failure of inhibitory control looks like from the outside, and the GABA-A receptor is the target of a large part of the anticonvulsant pharmacopoeia [69]. Temporal lobe epilepsy in particular involves changes in GABA and its receptors [70], and one of the harder clinical problems is seizures that stop responding to benzodiazepines at all [71].

Notice what that means about the drugs in the earlier sections. The compounds that take an hour of your memory are, chemically, close relatives of the compounds that stop a seizure. They are not two families. They are one family aimed at the same receptor, and the difference between a therapy and a side effect is largely a question of which subunits you hit and where.

Schizophrenia shows one of the most replicated findings in psychiatric neuroscience: abnormalities in cortical inhibitory neurons, particularly the parvalbumin-expressing type [72] [73]. Those are exactly the cells that generate gamma, which links the cellular finding to the rhythm abnormalities seen in the same patients [74].

Alzheimer's disease turns out not to be simply a loss of excitation. Mouse models show aberrant excitatory activity with compensatory remodelling of inhibitory hippocampal circuits [75], and reactive astrocytes releasing excess GABA impair memory in those models [76]. Too much brake, in the wrong place, from the wrong cells.

There is a pattern across these conditions that is easy to miss when they are listed one after another. In none of them is the problem simply too little inhibition. Epilepsy involves inhibition failing. Alzheimer's models involve too much of it, from the wrong cells, in the wrong place. Schizophrenia involves a specific class of inhibitory cell working badly while the rest carry on.

"Low GABA" is not a diagnosis, and any explanation that reduces to it is skipping the part that matters.

Depression has its own GABAergic deficit hypothesis with real support behind it [77], and GABA-A modulators are being actively developed as antidepressants [78].

The autism case is the one to state most carefully. An influential 2003 proposal framed autism as an increased ratio of excitation to inhibition in key systems [79]. That is a model. It has been productive, it has generated a great deal of research, and it is still argued about. Presenting it as a finding would misrepresent where the field is.

Even the receptor's own regulation is part of the story. How many receptors a synapse holds is itself controlled, and that trafficking is a form of plasticity in its own right [80]. GABA even guides how newly generated neurons integrate into adult circuits [81], which is a strange job for a brake and a good reminder that these labels are ours, not the brain's. Whether we will ever build drugs precise enough to target one subtype and leave the rest alone is still open [82].

The Pill That Shares a Name

One question deserves a straight answer, because it is asked constantly and answered badly almost everywhere.

You can buy GABA in a bottle. Does swallowing it do what the neurotransmitter does?

The problem is not whether the molecule is real. It is whether it gets where it would need to be. GABA the molecule crosses the blood-brain barrier poorly, and that permeability has been measured directly rather than assumed [83]. A molecule that cannot reach the receptor cannot act on the receptor. Oral GABA has nonetheless been studied for effects on anxiety-related behaviour, and some work reports them [84]. What is worth noticing is that the mechanisms proposed in that literature do not generally require the molecule to arrive in the brain intact. There is a lot of GABA signalling in the gut and the peripheral nervous system, and effects routed through there are a different claim from the one on the label.

So: the supplement and the neurotransmitter share a name. Sharing a name is not the same as sharing a route. That is the whole honest answer, and it is shorter than the pages that avoid giving it.

What Any of This Is Good For

Come back to the room one last time.

Which leaves the question of what you do with any of this, and the honest answer is not very much, practically. That is fine. Not every piece of knowledge about your own body has to convert into an action.

You were awake. You talked to someone. Your hippocampus was intact, undamaged, and fully capable of forming a memory. What had changed was that a drug had bound a site next to the GABA site on a subset of your GABA-A receptors, made them more responsive to the GABA your own interneurons were already releasing, and increased the inhibitory load on a structure where one particular subunit happens to concentrate.

The hour is not lost. It was never taken down.

That distinction is not wordplay, and it is the thing worth carrying away from all of this. It changes what you think happened to you.

It explains why the gap has a hard edge at the moment the drug took effect and why everything before it survived. It fits with rats getting better at learning when the same receptors are pushed the other way. And it explains why, if you ask someone about that hour, they will sometimes give you a confident account of things that never occurred, which is not lying but a mind doing what minds do with a blank.

It also changes what you should expect from anyone who was in the room with you. The nurse who spoke to you remembers the conversation. You are not being told about a memory you have mislaid. You are being told about an event you attended and did not record, which is a stranger thing to be on the receiving end of and worth naming accurately.

There is a broader point underneath. We describe inhibition as the negative half of the brain, the part that stops things. What the last fifty years of this literature actually shows is that inhibition is where a lot of the computation lives. It sets the gain. It generates the rhythm. It decides which of a thousand simultaneous inputs gets to matter. Delete a piece of it from a mouse hippocampus and the mouse does not become simpler. It becomes different, and on some tasks it becomes better.

The brake is not the opposite of thinking. It is a large part of how thinking gets done. Flumazenil, the drug that blocks the benzodiazepine site, is used clinically to reverse sedation [85], and the fact that one small molecule can lift the whole effect is the cleanest evidence there is that a single site on a single receptor was carrying all of it.

An hour of your life, held in a pocket on a protein.