LUNAROPS · OPERATIONAL UPLINK 100% UPTIME 1,247d POSTS 893 JEFF.MOON@LUNAROPS.DEV UTC --:--:--

How Anesthesia Works (and Why We Still Can't Fully Explain It)

14 min read The Human Machine
anesthesianeurosciencepharmacologyconsciousnessmedicinegaba
Contents

Roughly 300 million people are put under general anesthesia every year, and for most of them it works so reliably that they treat it as a solved problem: you count backward from ten, you reach seven, and then you are in the recovery room with no memory of the intervening hours. It feels like a light switch. It is not. Anesthesia is a controlled, reversible, dose-dependent demolition of consciousness, achieved with drugs whose detailed mechanism we genuinely cannot fully explain. We can predict the dose. We can titrate to effect. We can monitor the brain well enough to keep accidental awareness rare. But if you ask a pharmacologist why a noble gas with no metabolism and no classical receptor binding can erase your subjective experience, the honest answer is that we have several partial theories and no complete one. This is the rare case in medicine where the practical engineering is far ahead of the underlying science.

Anesthesia is not one thing

The first misconception worth killing is that “anesthesia” is a single effect. General anesthesia is a bundle of separable end states, and modern practice deliberately produces each one with different drugs acting on different targets. The classic teaching is the anesthetic triad, now usually expanded to four components:

Component What it means Failure mode if absent
Hypnosis (unconsciousness) No subjective experience, no awareness Awake but possibly unable to signal
Amnesia No memory formation of the period Explicit or implicit recall of surgery
Analgesia Blunting of nociceptive (pain) signaling Hemodynamic surges, stress response
Immobility / areflexia No movement in response to incision Patient moves during surgery

These are dissociable. Ketamine can produce profound analgesia and a dissociative state while the patient’s eyes are open and they are partly “aware.” Nitrous oxide gives analgesia and amnesia at concentrations that barely touch consciousness. Neuromuscular blockers like rocuronium produce total immobility while contributing nothing to unconsciousness or amnesia — which is exactly why paralysis without adequate hypnosis is the nightmare scenario of intraoperative awareness. A modern “balanced” anesthetic is therefore a recipe: a hypnotic (propofol or a volatile agent) for unconsciousness and amnesia, an opioid (fentanyl, remifentanil) for analgesia, and a muscle relaxant for immobility, each dosed against a different target. There is no single molecule that cleanly does all four, and pretending there is leads directly to the failure modes in the right-hand column.

MAC: the closest thing to a dosing ruler

For the inhaled (volatile) agents, the field has a remarkably durable unit of potency: MAC, the minimum alveolar concentration. Defined by Ted Eger and colleagues in the 1960s, 1 MAC is the alveolar concentration of an agent, at one atmosphere and steady state, that prevents movement in response to a standardized surgical incision in 50% of patients. It is, in effect, an ED50 expressed as a gas concentration — and because alveolar partial pressure equilibrates with brain partial pressure, it is a usable proxy for how much drug is actually at the target.

Agent MAC (vol % at age 40) Blood:gas coefficient Notes
Nitrous oxide ~104 0.47 Cannot reach 1 MAC alone at 1 atm
Desflurane ~6.0 0.42 Low solubility, fast on/off
Sevoflurane ~1.8 0.65 Standard for inhalational induction
Isoflurane ~1.15 1.4 Cheap, slower kinetics
Halothane ~0.75 2.4 Largely retired
Xenon ~63–71 0.115 Inert noble gas, yet anesthetic

A few things in that table matter more than they look. MAC is roughly additive: 0.5 MAC of nitrous plus 0.5 MAC of sevoflurane gives you about 1 MAC of effect, which is the entire basis of combining agents. MAC also shifts predictably with physiology — it falls about 6% per decade of age, drops with hypothermia, opioids, and pregnancy, and rises with chronic alcohol use and acute stimulants. And the concentration that abolishes consciousness is lower than the concentration that abolishes movement: MAC-awake, the concentration at which patients open their eyes to command, sits around 0.3–0.5 MAC. That gap is a clue we will return to — movement is suppressed largely in the spinal cord, while consciousness is suppressed in the brain, and they are not the same dial.

For intravenous agents like propofol there is no alveolar concentration to measure, so practice shifted to target-controlled infusion (TCI): pharmacokinetic models (Marsh, Schnider) estimate the effect-site concentration in real time and drive the pump. The effect-site concentration for loss of consciousness with propofol is roughly 2.5–3.5 µg/mL, and a typical induction bolus is 1.5–2.5 mg/kg. The model is doing the same job MAC does for gases: turning “how deep is this patient” into a number you can titrate against.

Meyer-Overton, and a century chasing the wrong target

The oldest theory of anesthesia is also the most seductive, because it is a clean correlation that held for a hundred years. Around 1900, Hans Meyer and Charles Overton independently noticed that the potency of an anesthetic tracks its lipid solubility almost perfectly: plot the oil:gas partition coefficient against 1/MAC and you get a straight line spanning four to five orders of magnitude, from feeble nitrous oxide to potent halothane. The obvious interpretation was that anesthetics dissolve into the lipid bilayer of neuronal membranes and disrupt them by sheer bulk — a non-specific, physical mechanism with no particular receptor.

   potent
   1/MAC
     ^
     |                                          . halothane
     |                                    .
     |                              . isoflurane
     |                        .
     |                  . sevoflurane
     |            .
     |      . desflurane
     | .
     |. nitrous oxide / xenon
   weak +-------------------------------------------> 
        low      oil : gas partition coefficient      high
        (Meyer-Overton: potency tracks lipid solubility)

The correlation is real. The interpretation was wrong, and the way it was disproven is one of the cleaner detective stories in pharmacology. Three facts broke the lipid theory. First, the cutoff effect: within a homologous series of n-alkanols, potency rises with chain length exactly as lipid theory predicts — right up until a certain length, where it abruptly vanishes even though solubility keeps climbing. A pure dissolving mechanism has no reason to have a cutoff; a binding pocket of finite size does. Second, stereoselectivity: the two mirror-image enantiomers of isoflurane and of etomidate have different anesthetic potencies despite identical lipid solubility. A lipid bilayer cannot tell left hand from right; a chiral protein binding site can. Third, in 1984 Franks and Lieb showed that anesthetics inhibit a pure soluble protein — firefly luciferase — at the same concentrations at which they cause anesthesia, with no membrane in sight. Anesthetics, in other words, bind proteins. Meyer-Overton survived not because anesthetics love lipids but because hydrophobicity is what lets a molecule reach a greasy hydrophobic pocket inside a membrane protein. The correlation was pointing at the right neighborhood and the wrong house.

The receptors we can actually name

Once the target became “proteins,” the field could finally start naming them, and the dominant story is about ion channels — specifically the balance between inhibition and excitation in the brain. Most general anesthetics tilt that balance toward inhibition, and they do it through a small set of channels.

Drug Primary molecular target Direction
Propofol GABA-A receptor (β subunit) Potentiates inhibition
Etomidate GABA-A receptor (β2/β3 subunit) Potentiates inhibition
Volatile agents (sevo, iso, des) GABA-A, glycine, TREK-1 K+, NMDA Mixed, net inhibitory
Barbiturates GABA-A receptor Potentiates inhibition
Benzodiazepines GABA-A receptor (γ subunit, classic BZD site) Potentiates inhibition
Ketamine NMDA receptor Blocks excitation
Nitrous oxide NMDA receptor Blocks excitation
Xenon NMDA receptor Blocks excitation
Dexmedetomidine α2-adrenergic receptor (locus coeruleus) Hijacks natural sleep pathway

The GABA-A receptor is the workhorse. It is a ligand-gated chloride channel; when GABA, the brain’s main inhibitory neurotransmitter, binds it, chloride flows in and the neuron hyperpolarizes — harder to fire. Propofol, etomidate, the barbiturates, the benzodiazepines, and a large share of the volatile-agent effect all potentiate this channel, either making it open more readily in response to GABA or, at higher concentrations, opening it directly. Etomidate is the cleanest example: it binds a well-characterized pocket at the interface of the β subunit, and engineered mutations at a single residue (β3-N265M in mice) make animals dramatically resistant to it. That is about as close to a smoking gun as molecular anesthesiology gets.

The NMDA receptor is the other pole. It is an excitatory glutamate channel, and the dissociative agents — ketamine, nitrous oxide, and xenon — block it. This is why those drugs feel different and monitor differently: they are not boosting inhibition, they are cutting excitation, and the resulting state is less “quiet brain” than “disconnected brain.” Then there are the leak channels: the two-pore-domain potassium channels like TREK-1, which volatile agents open to make neurons chronically leakier and harder to excite, and glycine receptors in the spinal cord, which are a big part of how volatiles produce immobility. Dexmedetomidine is the interesting outlier — it is an α2 agonist that works on the locus coeruleus and produces a state that overlaps with natural non-REM sleep architecture, which is why “sedation” with it looks more like rousable sleep than classical anesthesia. None of this is unlike the receptor-level story behind everyday neuroactive drugs; the same dose-response thinking that governs caffeine’s pharmacokinetics governs how an anesthesiologist titrates propofol, just with a far steeper consequence curve.

From molecules to a missing person: the network view

Here is the gap that keeps anesthesia interesting. Knowing that propofol potentiates GABA-A tells you why individual neurons go quiet. It does not tell you why you disappear. Plenty of things hyperpolarize neurons without abolishing consciousness. The leading modern view is that anesthetics abolish consciousness not by silencing the brain but by disrupting the integration of information across it — by cutting the long-range communication, particularly between the thalamus and the cortex and across frontoparietal networks, that lets separate brain regions act as one system.

Several lines of evidence converge on this. Functional imaging shows that as patients lose consciousness, the thalamus — the central relay that gates sensory traffic to the cortex — is consistently deactivated, and the functional connectivity between frontal and parietal cortex breaks down even when local activity continues. The most striking experiment uses TMS-EEG: pulse the cortex with a magnetic stimulus and record how the response propagates. In a waking brain, the perturbation spreads in a complex, differentiated pattern across regions. Under propofol, the same pulse produces a simple, local, stereotyped response that dies out quickly — the cortex still responds, but the regions can no longer talk to each other in a rich way. This is the basis of the perturbational complexity index and connects to integrated-information theory: consciousness seems to require a system that is both highly differentiated and highly integrated, and anesthesia knocks out the integration.

The practical upshot is that “depth of anesthesia” is really “degree of disconnection,” and the spinal immobility we titrate MAC against is a partly separate phenomenon happening below the brain entirely. That MAC-awake gap from earlier is the same idea seen from the dosing side: you can disconnect cortex enough to abolish experience at a concentration that does not yet abolish a spinal withdrawal reflex.

Watching a brain go under

Because the molecular and network stories are incomplete, monitoring is mostly empirical — we watch what the brain does rather than what the drug binds. The richest signal is the EEG, which changes in a stereotyped sequence as a patient goes under. As propofol takes hold, the EEG shows anteriorization (power shifting frontally), the emergence of large slow-delta and alpha oscillations, and, at deep planes, burst suppression: alternating periods of activity and near-flatline silence. Push deeper still and you reach an isoelectric, flat EEG.

 Awake     |~~~~~~~~~~~~~~~  low-amplitude, high-frequency, "busy"
           |
 Sedation  |/\/\/\/\/\/\/\   beta/alpha, slight slowing
           |
 General   |  /\    /\    /\  large frontal alpha + slow delta
 anesth.   |
           |
 Deep      |__/VVV\____/VVV\_  BURST SUPPRESSION (active bursts,
           |                    flat gaps; gaps lengthen with depth)
           |
 Too deep  |________________  isoelectric / flat

Reading raw EEG intraoperatively is a skill, so the industry built processed indices that crush the waveform into a single 0–100 number. The Bispectral Index (BIS) is the best known: 100 is awake, a target of roughly 40–60 indicates general anesthesia adequate for surgery, and below 40 suggests an unnecessarily deep, burst-suppressed state associated with worse outcomes. These monitors genuinely help titrate the GABAergic hypnotics. But they have honest limitations that every anesthesiologist learns the hard way. The proprietary algorithms were trained largely on propofol and volatiles, so ketamine and nitrous oxide can fool them — ketamine often raises the BIS number while the patient is profoundly anesthetized, because high-frequency EEG activity it produces looks “awake” to an algorithm tuned for the opposite drug class. EMG artifact from forehead muscles contaminates the signal. And the large outcome trials (B-Aware, B-Unaware, BAG-RECALL) disagree about whether BIS-guided care actually reduces awareness more than simply monitoring end-tidal agent concentration does.

That awareness risk is the reason any of this matters. Intraoperative awareness with explicit recall is rare but not mythical. Older volunteer-reported studies put it around 0.1–0.2% (one to two per thousand) when neuromuscular blockade is used; the large UK NAP5 audit, counting only patients who spontaneously reported it, found about 1 in 19,000 overall, rising in high-risk settings like cardiac surgery and cesarean section under general anesthesia. The common thread in awareness cases is almost always the same: adequate paralysis with inadequate hypnosis. Paralyze a patient and under-dose the propofol, and you have produced the one component (immobility) without the two that matter most (unconsciousness, amnesia) — the exact failure mode the four-component model warns about.

What we still cannot explain

For all of the above, the central mystery is intact. We do not have a unified theory that explains, from first principles, why this particular set of physically dissimilar molecules — a steroid-like IV drug, a halogenated ether, and an inert noble gas with a full electron shell and no metabolism — all produce the same reversible loss of consciousness. Xenon is the standing rebuke to every tidy story: it has no functional groups, undergoes no chemical reactions in the body, and is exhaled unchanged, yet it is a complete anesthetic. Whatever the deepest mechanism is, it has to be something a single inert atom can do, which rules out anything requiring chemistry and points back toward physical occupation of hydrophobic cavities in proteins.

The reversibility is its own puzzle. Consciousness comes back, intact and continuous, when the drug washes out — the demolition leaves no rubble. And the hardest question sits underneath all of it: anesthesia is the most reliable on-off switch we have for subjective experience, which makes it the best experimental tool we possess for studying consciousness, and yet we cannot say what is being switched. The network-integration view is the most promising frame, but it describes a correlate of consciousness, not its cause. We have learned to operate the switch with great precision while still not knowing what is on the other side of it — a state closer to surgical anesthesia than to natural sleep, despite the bedside language of “going to sleep” (the two are neurologically distinct; only dexmedetomidine sedation meaningfully resembles real sleep).

Verdict

General anesthesia is a triumph of empirical engineering sitting on top of incomplete science. The practical layer is genuinely solid: we have a usable potency unit (MAC) and pharmacokinetic models that let us titrate to effect, a clear map of the main molecular targets (GABA-A potentiation for most hypnotics, NMDA blockade for the dissociatives), a network-level account of why integration collapses, and EEG-based monitoring good enough to keep explicit awareness in the one-in-thousands-to-tens-of-thousands range. If you are a patient, the relevant fact is that anesthesiologists do not need the deep theory to keep you safe — the dose-response relationships are well characterized and the failure modes are understood. If you are an engineer or scientist, the relevant fact is the opposite: this is a domain where we can reliably and reversibly delete consciousness, dose it like a chemical process, and still not write down the mechanism. Meyer-Overton fooled the field for a century by pointing at lipids when the answer was proteins. It is worth holding some humility that our current network-integration story might be the next Meyer-Overton — a beautiful correlation that turns out to be aimed at the right neighborhood and the wrong house.

Sources

Comments