The drug fills your synapses with serotonin in hours, but people don't feel better for two or three weeks — that gap is where the actual mechanism hides.
"Feeling low" and "depression" are two different things, the way "out of breath after a run" and "asthma" are two different things. One is weather; the other is climate. It changes the parameters of the circuitry that regulates mood itself — so good news can't get in, bad news can't get out, and even the act of "looking on the bright side" stops working. This issue stays at the mechanism level: which circuits actually shift, which upstream levers you can still reach from outside, and one more urgent question — when you must take your hands off the levers entirely and get help.
Sixty years ago people noticed that a few drugs lifted mood, and that those drugs happened to act on a signalling molecule in the brain called serotonin (one of the chemical messengers neurons use to talk to each other). From that came a conclusion still repeated today: depression must be too little serotonin. The reasoning was faulty from the start — aspirin brings down a fever, which does not mean fevers are caused by an aspirin deficiency.
In 2022 a review pulling together decades of evidence answered plainly: there is no consistent evidence that depression is caused by low serotonin levels or activity. Metabolite measurements, receptor studies, experimentally stripping serotonin's dietary precursor out of healthy volunteers to see whether they crash — none of it gave stable support. As an account of the cause, "chemical imbalance" does not stand.
Yet almost simultaneously, another synthesis covering more than 500 trials reported that all twenty-one antidepressants beat placebo — the effect sizes aren't dramatic, but they're real. Both statements being true feels contradictory, until you look at the timing:
These drugs raise serotonin between neurons within hours. Patients typically need two to four weeks to improve. If the cause were really an empty tank, refilling it should help immediately — the way iron helps anaemia. So whatever heals is what happens during those weeks: sustained high serotonin drives a slow downstream program — BDNF (a protein that acts like fertiliser for neurons) goes up, synapses regrow, dendritic branches pruned back by chronic stress slowly come back. The drug presses the plasticity accelerator; it isn't topping up a tank. (How connections strengthen and weaken: synaptic plasticity)
Ketamine makes the same point from the other direction: it bypasses serotonin, pushes synaptogenesis straight through the glutamate line, and can lift severe depression within hours — fast precisely because it plugs in closer to the "regrow connections" step.
If it isn't a fluid level, what is it? The mainstream answer today sits at the circuit level: several parts that normally hold each other in check drift out of balance.
The alarm is too sensitive — the amygdala responds more strongly to negative input and settles more slowly (amygdala). The brake is loose — top-down regulation from prefrontal cortex weakens, so "I know I shouldn't think this" fails to hold (prefrontal cortex). One switch is stuck on — a small midline region, the subgenual anterior cingulate (area 25), runs persistently hot and cools down when treatment works; it's the classic deep brain stimulation target, though the boundary matters: early small open-label trials looked striking, the formal large controlled trial missed its primary endpoint, so don't file this as settled. Rumination won't stop — the default mode network (the one most active when you're staring into space, default mode network) is harder to exit and its content skews self-referential: not "what do I do about this" but "what is wrong with me."
Retold in the language of Topic 1, it gets clearer: depression looks like a hardened prior. The model "nothing I do matters" is weighted so heavily that evidence can't move it — and once it holds, you stop trying; stop trying and you never sample a counterexample, so the model is never overturned. That loop explains "why reassurance doesn't work" far better than any neurotransmitter does.
The loop has an exact counterpart in AI. One mainstream way to train agents is reinforcement learning: try things repeatedly, update your estimate of "is this worth doing" from how they turn out. It has a famous failure mode — an action happens to come out badly a few times in a row, its estimated value gets pushed low, and the agent never picks it again; never picking it means never receiving the counterexample, so the wrong low estimate is frozen in place. Meanwhile the old psychology experiment on "learned helplessness" was rewritten by one of its own authors: passivity turns out to be the default, and what animals actually learn is that they have control — a signal coming from prefrontal cortex, which, once it detects control, reaches down and suppresses the brainstem's default passivity switch. Both sides point at the same exit: you don't thaw it by arguing, you thaw it by manufacturing one real, small experience of control. That's the mechanism behind why behavioural activation starts with the tiniest possible action.
The circuit is hard to touch directly, but three upstream lines are reachable, each with a clean pathway.
One: the stress axis. Under threat the hypothalamus signals, the pituitary relays, the adrenal glands release cortisol — a three-stage cascade called the HPA axis, extremely useful in the short term. The key feature is that it's supposed to shut itself off: risen cortisol reaches back and presses the upstream switches down. Chronic stress blunts that brake, cortisol stays elevated, and the hippocampus — a key station enforcing the brake — is itself unusually vulnerable to cortisol. So: blunter brake → higher cortisol → blunter brake. (the HPA axis)
Two: inflammation. Roughly a quarter of depressed patients have elevated peripheral inflammatory markers. Inflammation on its own produces a whole package of "sickness behaviour": withdrawal, loss of interest, sleepiness, heightened pain sensitivity — overlapping the depression symptom list to a startling degree. More interesting is what anti-inflammatory trials show: an effect in the subgroup whose inflammation was high to begin with, which washes out when you average across all depressed people. That's a valuable piece of information — it means depression isn't one illness but several upstreams draining into one endpoint.
Three: rhythm. Insomnia was long treated as a symptom of depression; the evidence now says it is also a cause. Insomnia first and depression later is measurable, and treating the insomnia specifically (with cognitive behavioural therapy, not sleeping pills) lowers the later incidence of depression. Of the three levers, it's also the easiest to get hold of.
Doses, durations and how to sequence any of this belong to health-longevity; here we only pin down which line each knob pulls. Exercise pulls the BDNF-and-hippocampus line (Topic 18). Regular timing and morning light pull the phase of the clock and the stress axis (Topics 22, 23). Talking therapy pulls the prefrontal brake on rumination.
The commonest misreading is to call a day of ups and downs "bipolar." Real bipolar disorder is episodic, measured in weeks rather than hours — a stretch of time in which the whole baseline is lifted or pressed down, then switches.
And the early sign with the most diagnostic value isn't happiness, it's reduced need for sleep: three or four hours, and the next day you're not tired — your mind feels faster, ideas come thicker, speech speeds up. Someone with ordinary insomnia feels awful the next day; here the telling contrast is that it doesn't feel bad. Inflating plans, spending, and risk-taking often ride along with it.
Sleep here runs both ways: losing it isn't only a consequence, it can push someone into mania. So the self-correcting negative feedback of ordinary insomnia flips sign and becomes runaway positive feedback.
So the non-drug priority in bipolar isn't "more exercise, more sunlight" — it's nailing the daily schedule down: waking, eating, going out and seeing people at times that stay fixed, an approach known as social rhythm therapy. Two more hard facts: bipolar's heritability is high (twin studies commonly land in the 60–80% range, well above depression's); and antidepressants unopposed are dangerous in bipolar, capable of triggering a switch into mania or speeding up cycling, which is why guidelines require them to be covered by a mood stabiliser. The same low mood, but the cost of walking in the wrong direction is completely different — which is exactly what makes "I'll just try something myself first" so risky here.
None of that is covered by "tune the schedule a bit more." Get professional help; if there is immediate danger, call your local emergency number or go to the nearest emergency department. (In mainland China, the national mental health support line is 12356; emergency services are 120.) How talking therapy works and how to choose it belongs to psychology — this site only handles the mechanism.