The endocannabinoid system

In an ordinary synapse the message runs one way: the presynaptic terminal releases a transmitter and the postsynaptic cell receives it. Endocannabinoids run the other way. When a postsynaptic neuron is strongly depolarised, it synthesises anandamide or 2-arachidonoylglycerol (2-AG) on the spot, and those lipids drift backwards across the synapse onto CB1 receptors on the terminal that just fired, telling it to release less ^[1].

This was settled in a remarkable cluster of papers in 2001. Physiologists had known for years about two puzzling phenomena, depolarisation-induced suppression of inhibition and of excitation, in which strongly exciting a neuron briefly quietened the inputs onto it. Three groups showed within months of each other that the messenger responsible was an endocannabinoid, working through presynaptic CB1: at hippocampal synapses ^[8], from depolarised postsynaptic neurons onto their terminals ^[9], and by suppressing presynaptic calcium influx at excitatory synapses onto cerebellar Purkinje cells ^[10]. One mechanism explained a set of results that had looked unrelated.

On demand is the load-bearing phrase. Dopamine and serotonin are synthesised in advance and held in vesicles waiting for a signal. Endocannabinoids cannot be stored, because they are lipids that would simply dissolve into the membrane; they are cut out of membrane precursors at the moment of need and cleared within seconds to minutes ^[1]. Everything else about the system follows from that. The signal cannot travel far, it cannot last, and it cannot be depleted in the way a vesicle pool can be.

One further distinction is worth keeping, because it explains why blocking the system is so different from stimulating it. Alongside these brief phasic bursts, the system maintains a slow background tone: a low, continuous level of endocannabinoid signalling that holds baseline excitability, mood and appetite where they are ^[2]. A drug that adds a burst is doing something quite unlike a drug that removes the baseline, which is exactly the lesson the rimonabant story taught later.

Two receptors carry almost all of this, and the cleanest way to hold them is by location. CB1 is a brain receptor first: one of the most abundant G protein-coupled receptors in the central nervous system, concentrated on presynaptic terminals, and responsible for the psychoactive effects of cannabis as well as much of its influence on appetite, memory and motor control. CB2 sits mainly on immune cells and on microglia once they are activated, where it modulates inflammation without producing intoxication ^[1].

Both couple through Gi/o proteins, and on a nerve terminal CB1 closes calcium channels and opens potassium channels, which is the physical mechanism behind less transmitter release. That is as far as this page goes on the receptors themselves; the subtypes, their signalling, the ligands that prefer one over the other, and the additional targets such as GPR55, TRPV1 and the PPARs are covered properly at cannabinoid receptors.

The one point worth repeating here is the reason cannabinoid pharmacology is unusually rich: the plant compounds are not the system's ligands. THC, CBD, CBN, CBG and the rest borrow machinery built for anandamide and 2-AG, and each borrows it differently ^[1][2].

The receptors are one third of the system. The messengers and the enzymes are the other two, and the enzymes are where most of the drug development has happened, because changing how fast a messenger is destroyed is a subtler intervention than switching a receptor on.

Anandamide was the first to be found, isolated from brain in 1992 and named for the Sanskrit word for bliss ^[3]. It is a partial agonist at CB1 and is present at low concentrations. 2-AG was identified three years later, in canine gut and independently in brain ^[4][5]. It is a full agonist and is present at roughly a thousandfold higher concentration than anandamide, which is why 2-AG is now understood to carry most of the retrograde signalling while anandamide behaves more like a tone-setter. Two messengers at one receptor, with different efficacies and different abundances, is a large part of why the system can do so many things.

The building enzymes match that division. Anandamide is generated from a membrane phospholipid precursor by NAPE-PLD, a phospholipase D characterised in 2004 ^[6]. 2-AG is generated from diacylglycerol by diacylglycerol lipase alpha, and mice lacking that enzyme lose retrograde suppression at the synapse entirely, which is about as direct a demonstration of necessity as this field offers ^[7].

The destroying enzymes are the drug targets. Anandamide is hydrolysed by FAAH, cloned in 1996 ^[11]. 2-AG is hydrolysed mostly by MAGL, which a comprehensive profiling study showed accounts for the large majority of brain 2-AG hydrolysis, with two smaller enzymes handling the rest ^[12]. Because each messenger has its own clearance enzyme, it is possible in principle to raise one without the other, which is the whole premise of the section below.

The appeal of blocking a degrading enzyme rather than activating a receptor is that it only amplifies signalling that is already happening. An agonist floods every CB1 receptor in the body at once; a FAAH inhibitor raises anandamide where anandamide is already being made. On paper it is the more elegant intervention, and it should produce benefit without intoxication.

There is a compelling piece of human evidence that the premise is sound. A patient investigated for unusual pain insensitivity turned out to carry a microdeletion in a FAAH pseudogene alongside a hypomorphic FAAH variant, with markedly raised circulating anandamide, near-absent pain, and a striking absence of anxiety ^[13]. A natural experiment in permanently elevated anandamide produced roughly the phenotype the pharmacology predicted.

The drugs have not reproduced it. PF-04457845, a well-behaved irreversible FAAH inhibitor, raised anandamide substantially in a randomised placebo-controlled trial in knee osteoarthritis and produced no meaningful analgesia; the trial was designed carefully enough that the failure is informative rather than ambiguous ^[14]. The same compound later succeeded at something else entirely: in a randomised placebo-controlled trial in men with cannabis dependence it reduced withdrawal symptoms and cannabis use ^[15]. Raising anandamide does something real; it is not analgesia.

And the field carries a genuine tragedy. BIA 10-2474, a different FAAH inhibitor, was given in a phase 1 study in Rennes in 2016. After repeated dosing at the highest level, one volunteer died and four others sustained serious neurological injury with deep brain lesions ^[16]. Later work indicated the compound inhibited several lipases beyond FAAH, so the disaster is generally read as off-target toxicity rather than as a consequence of FAAH inhibition itself. Either way it is the clearest available reminder that a mechanism which looks gentle on paper is not thereby safe.

One further result deserves its place because it is the system's best-known everyday role. The runner's high is popularly attributed to endorphins. In mice, the reduced anxiety and pain sensitivity that follow running were abolished by blocking CB1 and were not abolished by blocking opioid receptors ^[26]. The endocannabinoid system, not the opioid system, produced the effect in that model.

THC is intoxicating because it is a direct CB1 agonist, and one detail explains much of its unpredictability: it is a partial agonist. How large its effect is therefore depends on how many receptors are present and how much endocannabinoid tone is already flowing, which varies by brain region and by person and by how much cannabis someone has been using ^[2]. Two people taking the same amount are not having the same pharmacological experience.

CBD does almost the opposite. It barely activates CB1 directly, so it is not intoxicating; instead it behaves as a negative allosteric modulator at CB1, binding elsewhere on the receptor and making agonists like THC less effective at it ^[18]. Much of what CBD does probably runs through other targets entirely, which is covered at cannabinoid receptors.

CBD is also the one cannabinoid whose clinical evidence is unambiguous, and it is worth being precise about what that evidence covers. In randomised placebo-controlled trials it reduced convulsive seizures in Dravet syndrome ^[19] and drop seizures in Lennox-Gastaut syndrome ^[20], which is the basis of its approval as a prescription medicine for those conditions. That is a narrow, well-supported indication in severe childhood epilepsy. It is not evidence for the wellness claims made for over-the-counter CBD, which mostly rest on preclinical work and open-label reports.

The interaction that matters most in practice comes from the same trials. CBD inhibits several liver CYP enzymes, and in children given CBD alongside clobazam the active metabolite of clobazam rose sharply, with sedation as the visible consequence ^[21]. Raised liver enzymes were also seen in the epilepsy trials, particularly alongside valproate ^[19]. CBD is metabolically active enough to change the exposure of other drugs, and that is true of the supplement as much as the medicine.

For the broader clinical picture across all cannabinoids, the best single summary remains a large systematic review and meta-analysis: moderate-quality evidence supporting use in chronic pain and in spasticity, low-quality evidence for most other proposed indications, and a consistently increased rate of adverse events ^[22]. Nabilone and dronabinol are licensed synthetic cannabinoids for chemotherapy-induced nausea and appetite stimulation, which is a real but narrow footprint for a system this broad.

The ECS attracts more confident claims than its evidence supports, in both directions. Four are worth handling directly.

Turning the system down is not a free lunch. Rimonabant, a CB1 inverse agonist, was approved in Europe as a weight-loss drug. A meta-analysis of the randomised trials found real weight loss and, alongside it, a significantly increased risk of depressive disorders and anxiety ^[17]. It was withdrawn in 2008. This is the strongest evidence anyone has that baseline endocannabinoid tone is doing something necessary for mood, and it was obtained the expensive way.

Clinical endocannabinoid deficiency is a hypothesis, not a diagnosis. The proposal that low endocannabinoid tone underlies migraine, fibromyalgia and irritable bowel syndrome has been argued in detail and assembles suggestive findings ^[25]. It has no accepted diagnostic test, no established reference range, and no treatment trial designed to test it directly. It is a reasonable hypothesis that is regularly quoted as though it were an established condition.

Heavy cannabis use changes the system, and the change reverses. PET imaging in chronic daily smokers found CB1 receptors downregulated across cortex, and the downregulation largely resolved after about four weeks of monitored abstinence ^[23]. That is the mechanism behind tolerance in this system; see tolerance and dependence.

Cannabinoid hyperemesis syndrome is real and is counterintuitive. Long-term heavy use can produce cycles of severe vomiting relieved temporarily by hot bathing, in a system whose acute effect is antiemetic. The mechanism is not established, and the only reliably effective treatment identified in a systematic review is stopping cannabis ^[24]. A drug being antiemetic acutely does not make it antiemetic chronically.

The honest gaps are worth stating plainly. Nobody can measure a person's endocannabinoid tone in a clinically useful way. No selective CB2 agonist has succeeded in a trial despite twenty years of interest in one. Whether MAGL inhibition can raise 2-AG therapeutically without producing the tolerance and receptor downregulation that prolonged blockade causes in animals is unresolved. And the ECS is unusually hard to study in people precisely because of what makes it interesting: the messengers are made on demand, act locally and are gone in seconds, so a blood measurement is a poor proxy for what is happening at a synapse ^[1][2].