A Chemical Mind

About A Chemical Mind

Stories of our fascination with the Brain: from medical mysteries, great triumphs and cautionary tales, to great discoveries and tragic failures, conspiracy theories, technology, and more; hosted by Nicholas Kircher (Published every Tuesday AU Time) chemicalmind.substack.com (https://chemicalmind.substack.com?utm_medium=podcast)

Publisher

Nicholas Kircher

Category

history · science

Language

en

Explicit

No

First episode

3 Dec 2023

Latest episode

28 Sept 2026

Latest episodes

34 episodes in the feed.

1. 

   28 Sept 2026

   How NOT to design a brain

   My good friend Tommy Blanchard (https://substack.com/profile/112941115-tommy-blanchard) wrote a great article the other week, on the question “Are humans (and biological organisms) simply wetware machines?” He gave a really good look at it with illustrations and everything, highly recommend reading it here: I just wanted to add some more interesting - for me - context to this point. In fact, I kind of want to do a loosey-goosey, unscientific look at what I’ve learned about the utter chaos of how the brain is put together by evolution. I think it really illustrates just how incredible these systems are, but also how mind-bogglingly complex and seemingly illogical, and how it might compare to a brain if we designed it like we might a computer. Prepare thyself. So the boss comes in to the office one morning. He rushes over to your desk. “Nicholas!” he says, breathlessly. “I’ve had a revelation!” Uh oh. It’s never good when the boss has had a revelation. “Since the brain is just too complicated to figure out, we should design a new one from the ground-up. It will be modular, maintainable, and debuggable!” You convey your misgivings for such a project, but their mind is made up. You’re now project lead for Brain 2.0 What now? Well you might start by analysing the existing behaviour of the current brain. In software engineering, when needing to do this, we often study the behaviour of the “legacy” version to identify each discrete “capability”, “feature” or “flow”, and map it to its own objective. Brain 2.0 should at the very least match the existing “legacy” behaviour. To know if it’s working, we have to formulate test criteria which the legacy brain should pass consistently to guide our work. The end goal should be for Brain 2.0 to also pass all those tests consistently. An example set of functions and features for a brain might consist of: * Sensory input processing * Language processing * Autonomic systems * Threat detection * Motivation state management * Bodily resource management * Memory * Motor function …and so on. Under each of these “feature buckets” we’d have to break it down into constituent parts based on what we understand of each set of behaviours or needs. For instance, “sensory input processing” is a very broad set of things. It includes vision, olfactory, tactile, proprioceptive and auditory inputs. We’d have to break each of these down into discrete modules. I’m already looking at it like: “Create a base schemata for modelling input data from a sensory organ, each sensory organ then extends that base with its own variations and needs.” This way, we have a basis for shaping the data for all sensory organs, and we can even add entirely new ones arbitrarily. Each sensory organ could have its information passed through a single module which would route the information to the correct destination for processing. So let’s say we get all this working, and now sensory inputs are coming in to the brain and informing it about the environment in which it finds itself. Furthermore, we have magically made motor control work, so our Brain 2.0 can respond to its environment via control of a physical body. But we’re running into problems. Something we need to consider is how to turn this discrete sensory information into a unified world model, but before we even get to that, we’re faced with a more serious challenge: compiling a world model directly from sensor data is slow. It takes time for signals to travel up nerve fibres, and yet more time to get filtered and processed, making it too slow to be viable when rapid near-reflex responses are sometimes required to avoid danger. To solve for this, we can borrow an idea from the legacy brain: its predictive model. The real human brain doesn’t feed sensory information directly into consciousness. Instead, it runs a continuous on-going prediction of the world, and uses sensory data only as a comparison to measure how accurate the prediction was, and to make corrections when necessary. The prediction is able to run ahead of the sensory input processing, making it ideal for rapid responses to potentially volatile environments since it doesn’t have to wait around for the data. So that’s what we’ll do with Brain 2.0! We’ll include a module for that as well! Notice what we’re doing here? We’re designing architectures, circuits, modules, patterns, and structures. We’re grouping things into neat, discrete boxes. We’re wiring things up cleanly and logically. The real human brain is nothing at all like this. Despite the way we often talk about it as though it were a modular architecture, it’s not really. It is a chaotic mess, worse than some ancient decades-old banking code base which has never been refactored. Change one thing, and you get unexpected side-effects all over the place, in ways that make no sense at all. Some people call evolution “The Blind Watchmaker”; quite frankly I think that’s an insult to blind people and watchmakers. It’s more like the proverbial infinite monkeys on typewriters, except also they’re blind and mildly deranged, have a total of 3 fingers, and for some unknown reason become obsessed with pressing only 4 keys on the typewriter: A, C, G, and T (or sometimes U instead of T, but no one knows why) Said monkeys never get even close to Shakespeare. Now, the universe didn’t really know what to make of this when it saw the crap these monkeys were coming up with. So in some strange act of madness, it invented a mechanism to interpret these long streams of 4-letter nonsense, and associated each letter with a nucleotide molecule: Adenosine, Cytosine, Guanine, and Thymine (or Urasil, depending on the monkeys particular whim) It is truly a remarkable achievement, then, that from such building blocks came all we see before us. Indeed, the legacy brain is a billions-of-years-old project. Every time evolution has demanded a new feature, or even a bug fix, our infinite blind and semi-deranged monkeys almost never clean up the old mess. Instead, they simply patch over the top of it. Everything becomes a patch on top of another patch. They live and die by one golden principle: if it increases survival and/or reproduction, ship it immediately and patch it over and over and over again. In digital computers, we would have electrical signals crossing wires and circuit pathways, transmitting digital information: a stream of 1s and 0s. By interpreting the specific sequences of 1s and 0s in a known way, we can derive its true meaning. Each processor that receives data knows how to interpret these signals in that way. In the wetware brain, things began much more simply. The earliest neural networks are thought to have been made out of highly multifunctional generalist cells, which combined sensory functions, inter-cellular communication and muscle contraction all into one, similar to the epithelial muscle cells still found in the modern cnidarian. Communication between these cells was necessary for rapid adaptive injury and stress responses. When damaged, a cell would release a flood of cheap, abundant signalling molecules into its surroundings to trigger coordinated, defensive/repair responses in neighbouring tissue. It was a rather crude but effective method of volume-transmission, and a bit like a mesh-network of light switches; damaged cell releases a messenger chemical, chemical binds to receptor on a neighbouring cell, cell activates a prepared response, and in turn releases its own chemical messenger molecules to recruit other neighbouring cells, and so on. This rapid cell recruitment and activation strategy proved highly successful, and the blind monkey that happened to stumble upon this particular configuration with its 4-letter alphabet was made to breed a whole new lineage of infinite blind monkeys. Then the scope creep began. These ancient multifunctional cells began specialising, separating the responsibilities of sensory receptor and muscle contraction into their own distinct cell types. Because these newly specialised sister cells became physically separated, and the traditional chemical messenger approach used before was too slow for communicating over larger cellular distances, a new type of cell which would bridge the gap between sensor and muscle contractor was needed: thus, the electrically-active neuron was born. Complex functionality first evolved within localised systems. Instead of a central brain suddenly appearing, small groups of these proto-neurons would connect up together in various parts of the body to control specific local reflexes. The Carribean Box Jellyfish, for example, has no central processing for sensory inputs, despite having 24 eyes. Instead, it has 4 separate “sensory ganglia” dangling off its body, each one processing input from 2 high-fidelity camera-like eyes, and 4 simpler ambient-light-sensors. These ganglia don’t communicate with each other, and have no world model. They continuously fire a regular pacemaker signal which controls the contractions of surrounding musculature for swimming. When one ganglia recognises an obstacle in the visual data from its 6 connected eyes, it changes the rate of the pacemaker signal, causing it to steer away. Over time, these networks grew to span even longer distances, and as organisms became more complex and evolved more sophisticated limbs and musculature, there was an evolutionary need for coordination and synchronisation. The Starfish’s approach to this problem was very simple. Each arm kept its localised network of nerve cells with all their localised sensory processing, but added peer-to-peer networking via a broadcast ring. When one of the arms detects food, a signal gets sent out to the other arms and degrades over distance. So the nearest neighbour arms will get a strong impulse, while the arms further away will get a weaker or no signal. The starfish will then crawl towards the food. The successes kept coming, so evolution went and scaled this up to an absurd level. Enter: the Octopus. Octopi have around 500 million neurons (a small fraction of a humans 86 billion), yet are highly intelligent. Their central brain consists of only about 10% of all its neurons, while the optic lobes have around 30%. The rest is distributed across its many infinitely-flexible arms. In fact, the central brain doesn’t control the limbs at all. Not directly. Instead, it might make strategic decisions and set objectives - “lets hunt”, “run away”, or something more specific like “grab that target” - while the local nervous system in each arm determines exactly how the kinematics for moving that arm will work to achieve the objective. The brain doesn’t coordinate movements either; each limb talks directly with other limbs to coordinate movements, something called “self-organised embodiment”. The signals from sensors on each arm are also processed locally, deciding whether something is food, a rock, or dangerous, before sending that information to the central brain for a strategic decision. This decentralised architecture is especially advantageous in the Octopus, as each arm has infinite degrees of freedom, and the processing required to compute all the kinematics for all the arms in one place would be overwhelming. Thanks to this decentralisation, in theory an octopus could have any number of additional arms added to it, without needing to change anything about the central brain. Interestingly, in software engineering, decentralisation has always been held up as a pillar of “good architecture.” We like the idea of being able to swap out components at will, and allow the system to reorganise itself according to certain rules. Change to something at a lower level of abstraction should not require major changes to things at higher levels of abstraction, and vice versa. In fact, it seems that invertebrates in general tend to follow this more decentralised architecture in their neural networks, while vertebrates opted for heavily centralised architectures where everything is hard-wired into the central brain, which micro-manages all downstream activity. You might, like me, wonder why. Wouldn’t this be more brittle? It is. We know this because of conditions like phantom limb syndrome. The human brain has a rigid map of the body hard-wired into itself, which cannot handle changes in configuration. This map is called the “cortical homunculus”. If a limb or other body part is removed or is disconnected from the brain or body, the neurons representing that body part in the homunculus can end up searching for other nearby neurons to connect to instead. Due to this rewiring, some amputees can feel sensations in their phantom limb when touching an area of the body adjacent to the phantom on the cortical homunculus; for example, due to their proximity, some people with foot or leg amputations can feel sensations in their phantom when their genitals are touched. (Neurologist V.S Ramachandran speculated this proximity might explain foot fetishes) Octopi have no such hard-wired body map. There are, however, some advantages to having such a rigidly centralised brain, which have served us well. The big headline one is imitation learning and mimicry. Humans are really good at “monkey see, monkey do.” Since the central brain has that unified knowledge of where each part of our body is located in physical space, our visual system can be presented with a shape and map it onto the cortical homunculus. So when we see someone else move in a certain way, we can mimic them pretty much immediately, something an Octopus simply cannot do as effectively. Over time, certain branches of the evolutionary tree really pushed hard on the brain centralisation track, and new features began to proliferate. If you look closely, you can see an interesting pattern forming: there seems to be inhibitory systems everywhere, apparently bolted on as an after-thought. For example, signals between two neurons can only flow in a single direction for a given synapse; there is almost always a strict division of labour between who the sender is and who the receiver is. However, at some point, evolution seemed to discover a need for the receiving neuron to have some control over how much signal it is being sent, so although it can’t send signals back in the other direction, it can do something called “retrograde inhibition”, telling the sender neuron to essentially “shut the fuck up” by throwing endocannabinoid molecules at it. In fact, this inhibitory bolting-on seems to happen at every level of abstraction. Lets take a look at something called the “direct pathway”, a bunch of ancient structures located deep in the brain which are heavily involved in voluntary motor movements. Even when we’re at rest, the motor cortex is continuously trying to get us to move, firing off signals non-stop for every possible movement we could make. The reason we’re not in constant motion at all times is thanks to a structure in the basal ganglia acting as a gate; by default, it fires a constant inhibitory signal to block motor movements. When we decide to move an arm to reach for a coffee cup, the cortex talks to the striatum, and the striatum sends a selective inhibitory signal to the gate, causing it to stop inhibiting those necessary motor signals. It inhibits the inhibitor. Then there’s the “indirect pathway”, which adds an extra inhibitor that inhibits the inhibitor of the inhibitor! It’s inhibitors all the way down!! It’s like if you were writing to a friend to tell them what colour the sky was, and so you write: “The sky where I am is not not blue, what colour is it where you are?” and your friend replies: “The sky here is not not not green.” In fact, our entire prefrontal cortex, one of evolutions most recent additions to the brain, is basically a giant slab of inhibitory neurons, helping to moderate and modulate our base impulses of fear, aggression, and drive for instant gratification. Without it, we would do the very first thing that popped into our heads moment to moment. Now, I’m not one of those militant atheists determined to argue the point, but I will say that perhaps those folks posting youtube videos about how a flagellum is evidence of intelligent design should take a closer look at the brain. The last point I’m going to make shows how truly incredible all of this is, because despite the sheer chaos and utter lack of any design whatsoever, the efficiency and power of biological neural networks are truly astounding. One of the most challenging things when it comes to actually simulating the biological brain on a computer is how much raw computational power it seems to require. Let’s compare, say, simulations for a fruit fly vs a mouse brain. So the fruit fly is only around 150,000 neurons and 50 million synapses, and if you run it as a simple stripped-down leaky integrate-and-fire network - like a typical AI model - you could run it comfortably on a laptop. However, to run a full simulation of the brain’s biology at real time speed would require about 100 Teraflops of computing power from a cluster of modern GPUs all running in parallel. You could technically do it with a home rig, if you had a motherboard with enough ports and data transfer bandwidth. It would require a ton of electricity, as well as making a lot of noise and heat, since the sheer communication overhead of tracking 50 million synapses forces us to throw massive amounts of hardware at the problem just to prevent the memory bottlenecks from slowing the simulation down. A mouse brain is around 71 million neurons, and 100 billion synapses; the neuron count is nearly 500 times larger, with 2,000 times more synapses. You couldn’t simply add more GPUs to your home cluster, because the bottleneck is in transferring the neuron data in these enormous matrices between all the separate compute devices, and then having the new data be re-integrated again at each step. The extra communication overhead for this would utterly overwhelm any cloud data centre. So first, you would need to obtain an enormous empty warehouse, like an aircraft hangar. Then you’d need to fill it with the very latest, most specialised, most powerful unified hardware. If we take for example the NVIDIA GB200 NVL72, which is essentially a rack-mounted 72-core GPU supercomputer on a chip, we get about 2.88 petaflops per unit. It has fully integrated memory, so inter-chip communication won’t become bottlenecked as easily. We’d likely need a full 200+ petaflops - with a P - to make this work. That means we’d need at least 70 racks of these. Each NVL72 draws 120kW of power. Running 70 of them at once would require 8.4 megawatts of electricity, continuously. But hey, at least you’d have one of the most powerful supercomputers that exists on this planet today. Which is kind of crazy, considering the biological mouse brain uses only about 8 milliwatts of power. The computational simulation of it requires over 1 billion times more energy to achieve. How in Darwin’s name did these infinite, blind, semi-deranged monkeys pull off such absurd levels of energy efficiency? Each neuron is like a single processor core operating in parallel. The fruit fly brain has 150,000 of these. That’s likely more parallel processing power than every single electronic device within a 1km radius of your house combined. The flip side is that neurons are not capable of everything that a single electronic CPU can do (which is a lot), however they don’t need to be. A lot of computational math is also obtained for free just by the nature of the organic chemistry happening at the cellular and synaptic layers. Digital hardware burns massive amounts of power to do large scale mathematics using rigid, binary logic gates. Evolutionary biology does it using fluid dynamics. The sprawling branches of a neuron act like microscopic, saltwater-filled cables with little holes that can open and close. When a synapse fires, it opens a hole at one end, letting a wave of voltage carried by ions rush in. The membrane is slightly leaky, so this wave naturally decays as it travels, giving the brain spatial filtering for free. If multiple waves fire in quick succession, they physically stack on top of each other in the fluid volume before leaking out, which gives the brain temporal integration. They leverage the natural electrical resistance of the cell membrane and the passive diffusion of ions, with positive and negative voltage ripples crashing into each other at the junctions of these branches, physically adding and cancelling each other out in real time. It gets complex calculus practically for free just by letting physics sort it out. Lastly, no central integration is required for biological neurons. They get their inputs, produce their outputs, and it just gets sent to the next set of connections. Each neuron can add or reconfigure or remove synaptic connections they have with other neurons in response to this, so this serves as fully integrated weighted memory. A computer, however, needs to keep shuttling the outputs from processors over to memory storage and back again to use as inputs just to simulate synaptic input/output. It’s a lot. All this is to say that while we can technically simulate biological brains on computers, it’s the many differences between logical design and evolutionary free-for-all which make it challenging. While we can use analogies to computers or modular architectures and the like to help us learn about the brain, we should do so with eyes wide open to the fact that it obscures just as much as it reveals. That’s all for today, thank you so much for reading all the way to the end. I hope I’ve been able to inspire you in some way with new thoughts, ideas and possibilities. If so, please consider upgrading to a paid subscription if you haven’t already, or drop me a donation over at ko-fi: I’ve been working hard on content for a complementary Youtube channel where I hope to deliver more digestible video versions of this stuff, as soon as the first set of episodes are ready, you’ll all be the first to know. Many thanks to everyone who has a subscription or made a donation at any point, you literally keep this show on the road. Until next time! Get full access to A Chemical Mind at chemicalmind.substack.com/subscribe (https://chemicalmind.substack.com/subscribe?utm_medium=podcast&utm_campaign=CTA_4)

2. 

   9 Jun 2026

   The Myths of Serotonin & Depression

   TLDR: Serotonin has no association with depression whatsoever. Instead, some cases are thought to be caused by reduced neuroplasticity during chronic periods of stress (mediated by the HPA axis and cortisol), resulting in a reduction in BDNF (brain-derived neurotrophic factor). SSRIs might work not due to serotonin, but due to significantly increasing BDNF expression, which is likely why psychedelics can also help in some cases. Few molecules in the history of biology have travelled as strange a road as serotonin. It was discovered twice, on two continents, by researchers chasing two completely different problems; one studying why blood serum constricts vessels, the other studying a substance in the gut that makes smooth muscle contract. Within a decade of its chemical identification it leapt from a curiosity of vascular physiology into the centre of psychiatry, becoming the basis of a “chemical imbalance” myth that has shaped how hundreds of millions of people think about depression. Today, that simple story has collapsed under the weight of evidence, even as serotonin has turned out to be far more biologically important - and far more widely distributed through the body - than its early champions ever imagined. In this episode, we trace the rise and fall of the “serotonin hypothesis” of depression, dig in to its surprisingly broad function that modern research has revealed, and take a peek at the current leading theories for what makes clinical depression happen (oh, and why SSRIs work at all!) Get in on these insights early! Hit the button: In the late 1930s, the Italian pharmacologist Vittorio Erspamer, working in Pavia, was studying a substance concentrated in the enterochromaffin cells of the gastrointestinal mucosa which caused smooth muscle to contract. He named it enteramine. Over years of painstaking work he and his colleagues characterised its biological actions, and Erspamer correctly suspected it was an “indole”-derivative (an unpleasantly-oderous organic compound found in coal tar and poo.) Interestingly, the Italians also played a major role in the West’s discovery of Dopamine, but that’s a whole other story (https://chemicalmind.substack.com/p/it-all-began-with-a-faba-bean). Independently, at the Cleveland Clinic in the United States, a team interested in the vasoconstrictor activity of blood serum was trying to isolate the agent responsible for raising vascular tone. Some 10 years later, in 1948, Maurice Rapport, Arda Green, and Irvine Page succeeded in isolating and crystallising the target agent from beef serum. Because it came from serom and affected vascular tone, they coined the name serotonin. Rapport then went on to determine its chemical structure, publishing the proposed vasoconstrictor principle the following year. Shortly afterward, serotonin was chemically synthesised, and it became clear that Erspamer’s enteramine and Rapport’s serotonin were the very same molecule: 5-hydroxytryptamine. This double discovery is why the molecule carries two naming traditions to this day: pharmacologists and neuroscientists usually write 5-HT, while the popular and clinical name remains serotonin. Then, in 1953, serotonin suddenly became a rather important subject of study to those who investigated diseases of the brain when Betty Twarog and Irvine Page demonstrated that serotonin was present in mammalian brain tissue. This was the moment serotonin became a candidate neurotransmitter, and the timing couldn’t have been more extraordinary. It coincided almost exactly with the discovery that the powerful hallucinogen lysergic acid diethylamide (LSD) was structurally related to serotonin, and could even antagonise (promote) its actions on smooth muscle (as serotonin does.) The inference - that a serotonin-like compound could so profoundly alter perception and mood, and make you hallucinate like the devil - electrified the young field of biological psychiatry and planted the seed of an idea that serotonin governed states of mind. Even before receptors could be cloned, classical pharmacology hinted at serotonin’s complexity. In 1957, Gaddum and Picarelli, studying guinea-pig ileum, proposed that serotonin acted on at least two distinct receptor types, which they called the “M” (morphine-blocked) and “D” (dibenzyline-blocked) receptors. This early two-receptor scheme was the ancestor of what is now recognised as one of the most complicated receptor families in all of pharmacology, encompassing at least 14 distinct receptor subtypes today. The serotonin theory of depression did not emerge from a direct observation that depressed people lacked serotonin. It emerged, somewhat backwards, from pharmacology. In the 1950s and 1960s, clinicians noticed that certain drugs altered mood, and researchers reasoned backward from the drugs’ known effects on brain chemistry to a presumed cause of the illness. Two foundational papers framed the debate: Joseph Schildkraut proposed the catecholamine hypothesis of affective disorders in 1965, arguing that depression might be associated with a deficiency of noradrenailne at functionally important brain sites. Two years later, in 1967, the British psychiatrist Alec Coppen advanced the case that 5-HT, rather than (or in addition to) the catecholamines, was central to the biochemistry of affective disorders. Together these papers crystallised what became the monoamine hypothesis of depression; the idea that mood disorders stem from a deficit of monoamine neurotransmitters in the brain, such as dopamine, noradrenaline, and serotonin. The hypothesis gained enormous traction because drugs that increase synaptic serotonin can apparently relieve depressive symptoms. The decisive commercial and cultural moment came with fluoxetine (Prozac), developed at Eli Lilly and described by David Wong and colleagues as the first selective serotonin reuptake inhibitor (SSRI) to reach the market. Wong’s retrospective review traces the deliberate, two-decade evolutionary process by which fluoxetine was engineered specifically to block serotonin reuptake while sparing other systems. SSRIs supposedly work by inhibiting the serotonin transporter (SERT), thereby raising the concentration of serotonin in the synaptic cleft. The marketing logic was seductive and simple: if a drug that raises serotonin treats depression, then depression must be caused by too little serotonin - a “chemical imbalance” that the medication corrects. This framing was widely communicated to the public through advertising and clinical encounters, and it became one of the most successful pieces of medical folk-knowledge of the late twentieth century. And, as Psychopharmacologist Stephen M. Stahl noted in his 1998 paper, Prozac and similar agents are “among the most frequently prescribed therapeutic agents in all of medicine.” It turns out, however, this simple pharmacokinetic story was built on far shakier foundations than anyone believed at the time. Crucially, the inference is a logical error of the form of “the drug raises X, therefore the disease is a deficiency of X.” Aspirin relieves headaches, but headaches are not caused by an aspirin deficiency. Even at the time, serious problems were visible, such as the therapeutic delay, and a heterogenous pharmacology of anti-depressants. SSRIs raise synaptic serotonin within hours, yet clinical antidepressant effects typically take weeks to appear. This temporal mismatch suggested that the relevant therapeutic mechanism is not the acute rise in serotonin itself, but slower, downstream adaptations. For example, one such adaptation is the desensitisation of somatodendritic 5-HT1A autoreceptors (these are receptors that detect the molecule in the “extra-cellular space”, i.e outside or overflowing the synaptic cleft) in the raphe nuclei. Meanwhile, drugs with very different effects on serotonin can all have antidepressant activity, which is hard to reconcile with a single, simple serotonin-deficiency model. Something was clearly not right with any of this. The drugs definitely work; we just couldn’t quite figure out how. The empirical case against the simple serotonin-deficiency model was assembled most comprehensively in a 2022 systematic umbrella review led by Joanna Moncrieff and colleagues, published in Molecular Psychiatry. Together, they synthesised the principal bodies of evidence: * serotonin and 5-HIAA (its main metabolite) concentrations in bodily fluids; * 5-HT1A receptor binding; * serotonin transporter (SERT) levels by imaging and post-mortem; * tryptophan-depletion experiments; * and SERT gene associations and gene–environment interactions. Their conclusions were revealing. First, meta-analysis of the serotonin metabolite - 5-HIAA - showed no association with depression. A meta-analysis of serotonin in blood plasma showed no relationship with depression, either. Shockingly, it actually found that lowered serotonin was associated with the use of antidepressants - suggesting these medications might be doing the exact opposite of what we all believed. Then, further analysis of the largest and highest-quality genetic observations of Serotonin Reuptake Transporter - SERT - showed no relationship whatsoever with depression, and no gene-by-stress interaction. Not only had we been wrong about how these medications worked, we had been ass-backwards wrong. Upside-down looney-tunes wrong. We had been living in opposite land. The bottom line was that the main areas of serotonin research provide no consistent evidence that depression is caused by lowered serotonin activity or concentration. The authors wrote that the areas surveyed “provide no consistent evidence of there being an association between serotonin and depression, and no support for the hypothesis that depression is caused by lowered serotonin activity or concentrations”; more provocatively still, “some evidence was consistent with the possibility that long-term antidepressant use reduces serotonin concentration.” (emphasis mine) The paper - as you might expect - generated intense debate; the journal published numerous commentaries both supporting and criticising its framing and methods, but it crystallised a scientific consensus that had in fact been building for years: The simple “low-serotonin-causes-depression” story is not supported by the evidence. It is important to be precise about what collapsed. The umbrella review undermined the claim that depression is caused by a serotonin deficiency. It does not prove that SSRIs are ineffective, nor that serotonin is irrelevant to mood. The efficacy of antidepressants is a separate empirical question, addressed by large network meta-analyses such as Cipriani and colleagues’ 2018 study of 21 antidepressants, which found that all examined antidepressants were more effective than placebo for acute major depression, albeit with modest effect sizes and varying acceptability. The most defensible modern position is that SSRIs can be clinically useful, but that their benefit does not validate a serotonin-deficiency theory of causation. Stahl puts it bluntly: “the immediate actions of SSRIs are mostly side effects,” and locates the cure elsewhere: “The explanation for therapeutic effects characteristic of SSRIs may be found in delayed neurochemical adaptations,” of which “a leading hypothesis… is desensitization of somatodendritic serotonin 1A autoreceptors in the midbrain raphe.” If the psychiatric story narrowed and then partially collapsed, the broader biology of serotonin expanded enormously. The single most important re-framing of the past two decades is captured in the title of a landmark 2009 review by Berger, Gray, and Roth: “The expanded biology of serotonin”. What we did in fact discover is that most serotonin is not located in the brain. Although serotonin is famous as a brain neurotransmitter, the overwhelming majority of the body’s serotonin is found outside the central nervous system, predominantly in the gastrointestinal tract, where it is produced by enterochromaffin cells and stored in circulating platelets. Within the brain, serotonin’s character is one of a neuromodulator via volume transmission, a characteristic it shares mainly with various kinds of peptide: it does not so much carry discrete point-to-point messages as set the gain and tone of large brain networks. Its functions are remarkably broad; it is involved in the regulation of mood, sleep, appetite, food intake, aggression, impulsivity, and many other processes, and it is the precursor for melatonin synthesis in the pineal gland. The diversity of serotonin receptor subtypes - each with its own distribution and signalling - is what allows a single molecule to influence so many distinct functions, and it explains both the therapeutic breadth and the side-effect profile of serotonergic drugs. So, if not serotonin deficiency, then what is depression? There’s a word - “stress” - that often comes up in the literature these days. I’ve often found myself wondering what exactly kind of “stress” is being referred to here. When we talk about stress, we’re often talking about psychological and psychosocial challenges, such as losing a job, concern over finances, marital instability, and so on. However, stress in biology is usually something different; infection, illness, disease, anything that results in systemic inflammation, in which large quantities of leukocytes (white blood cells) and cytokines flood the body, with wide-ranging biochemical impacts across all systems. Something that has emerged recently is an understanding that psychological stress doesn't stay psychological. It is converted into biological signals by the hypothalamic–pituitary–adrenal (HPA) axis, which releases cortisol. Sustained high cortisol is itself neurotoxic to the relevant circuits; it is one of the proximate drivers of synaptic atrophy, and a reduction in signalling of the brain’s growth hormone, “brain-derived neurotrophic factor” (BDNF.) It effectively slows your brain’s ability to adapt, significantly reducing its neuroplasticity, as well as weakening existing connections. Presently, this is hypothesised to be one of the leading causes of clinical depression: the inability for the brain to adapt away from negative thought patterns, due to the massive reduction in neuroplasticity caused by chronically high cortisol. Interestingly, this also seems to be where SSRIs have their anti-depressant effects; one idea posited that the chain of events that causes the desensitisation of serotonin autoreceptors in the brain may also result in the increase of BDNF expression, restoring levels of neuroplasticity that resemble the “critical period” of brain development. A related and increasingly influential idea is that serotonergic drugs (and psychedelics) may enhance a window of plasticity that allows the brain to “relearn” healthier patterns, with environment and psychotherapy determining the outcome. In this model, the SSRI unlocks the gate, allowing the patient to pass through. This would explain why combined SSRI and psychotherapy treatment regimens are typically much more successful than either one on its own: SSRIs seem to give the brain the opportunity to break out of negative thought pathways. These mechanisms are areas of intense ongoing investigation and are not yet settled. Although the leading hypothesis of depression has crumbled beneath our feet, the effect on research has been to blast wide-open many new and exciting areas that were once thought to be settled. The failure to find a robust SERT-gene association with depression is but one, showing depression is almost certainly not one disease but a heterogeneous collection of conditions with overlapping symptoms and many contributing causes, including genetic, developmental, inflammatory, social, and environmental. Identifying biologically meaningful subtypes, matching them to mechanisms (serotonergic or otherwise), and translating that into clinical treatments, is among the very frontiers of biological psychiatry today. There’s also more being found on the psychological front, such as how antidepressant treatments “decrease the negative bias in the processing of emotionally salient information early in the course of antidepressant treatment, which leads to the clinically significant mood improvement later in treatment”, as described by Godlewska and Harmer. This model reflects “a change in the view of psychological and biological processes, from seeing them as separate to complementing one another.” What we are still discovering may be the most exciting part: how antidepressants and psychedelics actually reshape the brain, how the gut and its microbes talk to the mind, and how a single small indole built from a dietary amino acid came to touch so many corners of human physiology. Serotonin’s story is not finished. If anything, after seventy-five years, it is becoming more important than ever. References * The discovery of serotonin and its role in neuroscience (https://doi.org/10.1016/S0893-133X(99)00031-7)Whitaker-Azmitia PM | Neuropsychopharmacology | 1999 * The serotonin theory of depression: a systematic umbrella review of the evidence (https://doi.org/10.1038/s41380-022-01661-0)Moncrieff J, Cooper RE, Stockmann T, Amendola S, Hengartner MP, Horowitz MA | Molecular Psychiatry | 2023 * Serum vasoconstrictor (serotonin): the presence of creatinine in the complex; a proposed structure of the vasoconstrictor principle (https://doi.org/10.1016/s0021-9258(19)51208-x)Rapport MM | Journal of Biological Chemistry | 1949 * Two kinds of tryptamine receptor (https://doi.org/10.1111/j.1476-5381.1957.tb00142.x)Gaddum JH, Picarelli ZP | British Journal of Pharmacology and Chemotherapy | 1957 * International Union of Basic and Clinical Pharmacology. CX. Classification of Receptors for 5-Hydroxytryptamine; Pharmacology and Function (https://doi.org/10.1124/pr.118.015552)Barnes NM, Ahern GP, Becamel C, Bockaert J, et al. | Pharmacological Reviews | 2021 * The catecholamine hypothesis of affective disorders: a review of supporting evidence (https://doi.org/10.1176/ajp.122.5.509)Schildkraut JJ | American Journal of Psychiatry | 1965 * The biochemistry of affective disorders (https://doi.org/10.1192/bjp.113.504.1237)Coppen A | British Journal of Psychiatry | 1967 * Prozac (fluoxetine, Lilly 110140), the first selective serotonin uptake inhibitor and an antidepressant drug: twenty years since its first publication (https://doi.org/10.1016/0024-3205(95)00209-O)Wong DT, Bymaster FP, Engleman EA | Life Sciences | 1995 * Mechanism of action of serotonin selective reuptake inhibitors. Serotonin receptors and pathways mediate therapeutic effects and side effects (https://doi.org/10.1016/s0165-0327(98)00221-3)Stahl SM | Journal of Affective Disorders | 1998 * Comparative efficacy and acceptability of 21 antidepressant drugs for the acute treatment of adults with major depressive disorder: a systematic review and network meta-analysis (https://doi.org/10.1016/S0140-6736(17)32802-7)Cipriani A, Furukawa TA, Salanti G, et al. | Lancet | 2018 * The expanded biology of serotonin (https://doi.org/10.1146/annurev.med.60.042307.110802)Berger M, Gray JA, Roth BL | Annual Review of Medicine | 2009 * A unique central tryptophan hydroxylase isoform (https://doi.org/10.1016/s0006-2952(03)00556-2)Walther DJ, Bader M | Biochemical Pharmacology | 2003 * Patients with high-bone-mass phenotype owing to Lrp5-T253I mutation have low plasma levels of serotonin (https://doi.org/10.1002/jbmr.44)Frost M, Andersen TE, Yadav V, Brixen K, Karsenty G, Kassem M | Journal of Bone and Mineral Research | 2010 * Indigenous bacteria from the gut microbiota regulate host serotonin biosynthesis (https://doi.org/10.1016/j.cell.2015.02.047)Yano JM, Yu K, Donaldson GP, Shastri GG, Ann P, Ma L, Nagler CR, Ismagilov RF, Mazmanian SK, Hsiao EY | Cell | 2015 * The role of glutamate underlying treatment-resistant depression (https://doi.org/10.9758/cpn.22.1034)Kim J, Kim TE, Lee SH, Koo JW | Clinical Psychopharmacology and Neuroscience | 2023 * Psychedelic Psychiatry’s Brave New World (https://doi.org/10.1016/j.cell.2020.03.020)Nutt D, Erritzoe D, Carhart-Harris R | Cell | 2020 * Cognitive neuropsychological theory of antidepressant action: a modern-day approach to depression and its treatment (https://doi.org/10.1007/s00213-019-05448-0)BR Godlewska, CJ Harmer | Psychopharmacology | 2020 Get full access to A Chemical Mind at chemicalmind.substack.com/subscribe (https://chemicalmind.substack.com/subscribe?utm_medium=podcast&utm_campaign=CTA_4)

3. 

   28 May 2026

   The Myths of Dopamine

   I know you’ve heard of it. “Dopamine.” It’s supposedly a chemical in the brain that causes pleasure, right? I mean, everyone says so. Even Harvard (https://www.health.harvard.edu/mind-and-mood/dopamine-the-pathway-to-pleasure) says so, and surely we can trust Harvard, right? No, sadly, this most perpetuated of common wisdom is perhaps the most common medical myth of our time. If you thought dopamine had anything to do with feelings of pleasure, it’s time to set the record straight. To do that, we need to start at the beginning of the story; the story of how dopamine became so misunderstood. In a sterile room of a McGill University laboratory in 1954, a surgical mistake birthed a fascinating discovery. James Olds and Peter Milner were attempting to study the neural mechanisms of learning in rats, using a microelectrode embedded into the midbrain reticular formation. When it came time to test “rat number 34”, the electrode slipped, misaligning by a fraction of a millimetre, and embedding itself into the septal area. Like the others, this rat was placed into their operant conditioning box and given a lever which, when pressed, delivered a fraction of a volt through the electrode into the brain. Unlike all the others, however, this particular animal became utterly transfixed by this device. It pressed the lever again, and again, and again. Soon, it was pressing it thousands of times an hour. The rat ignored food, it ignored water, and it ignored females in heat. It would actively stimulate its own brain until it collapsed from exhaustion, and starved to death. Olds and Milner concluded they had stumbled upon the mammalian brain’s “pleasure centres.” The pathways radiating from these regions were rich in a specific neurotransmitter: Dopamine. Thus began one of the most persistent biological myths, one that has survived for over 70 years despite rigorous scientific falsification, embedding itself in the popular consciousness, and even today being repeated in academic research as an axiom that “everyone knows” without any citation or reference. Look again at that article from Harvard; do you see any citations? This idea - that dopamine was “the brain’s pleasure chemical” - was further crystallised in 1978, when researcher Roy Wise formulated the “Anhedonia Hypothesis” after observing that administering the dopamine receptor antagonist (blocker) pimozide to rats caused them to stop working for food. To the naked eye, the rats behaved exactly as if the food had been removed entirely. Wise argued that blocking dopamine systematically eroded the “hedonic impact” of the reward itself. In this paradigm, dopamine was pleasure incarnate; chemically turning off dopamine meant chemically disabling the capacity for enjoyment. It was a neat, beautifully simple and consumable theory that resonated deeply in the minds of scientists and laypeople alike, and helped to invent many of the most ridiculous pop-psychology fads we see today. That anhedonia hypothesis held for a decade, until neuroscientists Kent C. Berridge and Terry Robinson fundamentally broke everything about the model in 1989. Berridge reasoned that if dopamine was the quintessential chemical of pleasure, a rat stripped of all dopamine should exhibit a total absence of hedonic response. The question was: how do we really test “hedonic response” with confidence? Using the neurotoxin 6-hydroxydopamine (6-OHDA, or “oxidopamine”), Berridge selectively destroyed 99 percent of the dopamine neurons in the rat striatum. Predictably, the rodents became entirely aphagic (refusal or inability to swallow). They refused to seek out food and would rapidly starve unless they were artificially fed via gastric tubes. However, Berridge also introduced a novel metric originally designed by Grill and Norgren in 1978: the “taste reactivity paradigm.” It turns out, rats give off a remarkably consistent signal of receptivity to taste through their facial expressions. It’s so consistent, in fact, it can be used as an empirical measurement. Given something sweet and delicious, rats exhibit rhythmic tongue protrusions and relaxed facial muscles: the “positive” response. When given something bitter and yucky, they gape with mouth open wide, jaw dropped, and corners of the mouth retracted, followed by a sequence of physical aversive reactions: the “negative” response. When given something entirely tasteless, such as room-temperature water, only rhythmic mouth movements occur, without either the tongue protrusions of sweetness or gaping from bitterness: the “neutral” response. If dopamine really was the chemical responsible for pleasurable sensations, blocking or destroying dopaminergic neurons should result in the total elimination of the “positive” response to sweet or delicious-tasting substances. In fact, that’s exactly what Berridge had been expecting to find: that all taste reactivity would become either “neutral” or “negative,” regardless of what was given to the rats. So when Berridge manually pipetted a sweet sucrose solution directly into the mouths of these dopamine-depleted rats and recorded their facial micro-expressions in slow motion, the results were… confusing. The rats still exhibited the same rhythmic tongue protrusions and relaxed facial muscles as would be found in the unmodified control rats; the positive pleasure response. So then what, exactly, was dopamine-depletion doing to them? If they still experienced pleasure, why were they so... depressed? Berridge and Robinson spent the subsequent decade proving an idea: that dopamine does not mediate “liking” (hedonic impact, pleasure). Instead, it mediates “wanting”. They believed dopamine in the mesolimbic pathway supports an unconscious tagging system that makes reward cues fundamentally attractive, propelling the organism’s behaviour toward an objective. They called this “incentive salience.” The dopamine-depleted rats still liked and enjoyed sugar; they merely stopped caring for it one way or another. They stopped pursuing it. They no longer particularly wanted it. It turns out there is a distinct difference in the brain between liking something and wanting something. True pleasure, or “liking,” is actually fragile and remarkably localised in the brain. It is mediated by isolated “hedonic hotspots” predominantly driven by opioid and endocannabinoid signalling in the nucleus accumbens shell and ventral pallidum. By contrast, the dopamine-driven “wanting” system is massive, robust, and highly susceptible to sensitisation. This neurological divergence explains the bleak reality of severe addiction. As Berridge noted, the dissociation of wanting and liking means a drug addict can reach a neurobiological state where their mesolimbic system, driven by dopamine, is so heavily sensitised to a particular sensory cue - in this case, a drug - that they possess a desperate, overwhelming want for it, even if subjective tolerance is so high that they no longer like the experience of taking it. Furthermore, in 2025, Berridge was able to demonstrate conclusively that it is possible for an organism to want something that they severely dislike by directly manipulating the mesolimbic pathway with electrical stimulation: using this mechanism they were able to cause rats to feel an overwhelming motivation to bite on an electrified rod, causing them pain and discomfort. Clearly, we can like things we don’t want, and want things we don’t like. It has been scientifically proven that you both want and like articles just like this one; subscribe! Here’s another fact about dopamine you probably didn’t know: your eyes use dopamine to signal the brightness of light entering the retina. Your brains motor cortex uses dopamine to gate movement signals for fine motor control. In fact, dopamine is used in lots of completely different ways throughout the brain and body; it makes no sense whatsoever to pin any one function on it. Suggesting dopamine is the brain’s pleasure chemical is not only specifically wrong, it’s also wrong in general. It would be like suggesting radio waves are how software engineers communicate cat videos with each other. Sure, if you follow enough reductionist reasoning, you can reach a level where cat videos are indeed carried across the internet which sometimes includes radio waves, but would anyone describe “the internet” explicitly as a “system for transmitting images and videos of cats?” Don’t answer that. (On that subject, this is one of my earliest youtube videos:) Anyway, you get my point. Sometimes people watch other sorts of videos on the internet, too. Not me, though I’ve heard rumours of it happening. So, if dopamine really isn’t relevant to the phenomenon of pleasure, why has the myth been so persistent? My only guess is that it lets us quickly and easily explain away things like addiction, ADHD, depression, “youths” (get off my lawn.) It makes it easy to say that drug addicts are just out to get “high”, that it is a choice they make  - choosing to seek pleasure even to the detriment of all else  -  which they could change if they really wanted to, if they simply “took responsibility for themselves.” It allows us to pretend ADHD isn’t real, that such people need to “be responsible”, as if it’s all just a choice they make to pursue only “enjoyable things”, or that it’s only a childhood phase one grows out of: because, you know, being a “grown up” means you start making “good choices” and “doing the hard things” which is the antithesis of ADHD, right? It makes it easy to wave away depression as merely an absence of joy, to say “come on, it’s not that bad”, as though the cure were merely to realise that butterflies and rainbows exist, and allow all the joy to flow back into the brain just like that. These simplistic explanations may allow us to forget about the complexities of the world around us, and especially our own brains. We don’t like things being too complicated; part of being human is trying to simplify and explain things, even things we don’t understand. We also don’t like to believe anything which suggests a lack of human control, especially over our choices. To believe it’s all just some chemical reaction makes it seem like it’s not really a conscious choice, doesn’t it? (It isn’t) The illusion of the ethereal conscious self as being in absolute control over our physical bodies, like an external puppet-master, is indeed a very powerful and very useful one to our sense of a unified identity, despite the fact that our bodies are an amalgamation of uncountable numbers of other micro-organisms. What’s more, it’s no doubt comforting to believe our choices are made by some fully integrated and consciously aware entity, as though we design the choices we make with intent. We don’t. It’s all just chemistry. References: Positive reinforcement produced by electrical stimulation of septal area and other regions of rat brain. (https://doi.org/10.1037/h0058775) | J. Olds and P. Milner | 1954 | Journal of Comparative and Physiological Psychology, 47(6), pp. 419-427. “the electrical stimulus in the septal area has a positive reinforcing effect of a very high order” Neuroleptic-induced “anhedonia” in rats: pimozide blocks reward quality of food. (https://doi.org/10.1126/science.566469) | R.A. Wise, J. Spindler, H. deWit and G.J. Gerberg | 1978 | Science, 201(4352), pp. 262-264. “Pimozide appears to selectively blunt the rewarding impact of food and other hedonic stimuli.” Taste reactivity analysis of 6-hydroxydopamine-induced aphagia: implications for arousal and anhedonia hypotheses of dopamine function. (https://doi.org/10.1037/0735-7044.103.1.36) | K.C. Berridge, I.L. Venier and T.E. Robinson | 1989 | Behavioral Neuroscience, 103(1), pp. 36-45. “The persistence of normal taste reactivity argues against... an anhedonia... hypothesis” What is the role of dopamine in reward: hedonic impact, reward learning, or incentive salience? (https://doi.org/10.1016/s0165-0173(98)00019-8) | K.C. Berridge and T.E. Robinson | 1998 | Brain Research Reviews, 28(3), pp. 309-369. The incentive sensitization theory of addiction: some current issues. (https://doi.org/10.1098/rstb.2008.0093) | T.E. Robinson and K.C. Berridge | 2008 | Philosophical Transactions of the Royal Society B, 363(1507), pp. 3137-3146. Wanting what hurts: D1 dopamine neuronal stimulation in CeA is sufficient to induce maladaptive attraction (https://www.nature.com/articles/s42003-025-08944-6) | Nguyen and Berridge | 2025 | Nature Get full access to A Chemical Mind at chemicalmind.substack.com/subscribe (https://chemicalmind.substack.com/subscribe?utm_medium=podcast&utm_campaign=CTA_4)

Podcast Authority Score: 12 / 100

A composite of feed quality, social presence, YouTube performance and engagement. Read the methodology.

Quality

24

Social presence

0

YouTube

0

Engagement

0

Contact A Chemical Mind

Guest appearances

Books guests

Based on episode analysis; this does not confirm that the show is currently accepting guests.