‹

01. The Myth of the Broken Thermostat

EN

01. The Myth of the Broken Thermostat

A thermostat is a simple machine. It reads a temperature, compares it to a target, and switches something on or off to close the gap. For decades, the most common public explanation of depression borrowed exactly that machine. Somewhere in the brain, the story went, a chemical called serotonin sits at too low a level, the way a room sits too cold. An antidepressant tops it back up, the way a furnace fires. The metaphor is clean, it is memorable, and it is the reason millions of people have been told, in a doctor's office or a pill insert, that they have a "chemical imbalance." In 2022, a team of researchers pulled the thermostat apart to see whether the wiring was really there.

The idea has a name and a lineage. It traces to a hypothesis associated with the psychiatrist George Coppen, and it hardened over fifty years into something close to public common sense: low serotonin, low mood. Serotonin itself is a signaling molecule, a chemical that nerve cells release to talk to one another across the tiny gaps between them, called synapses. It is involved in mood, sleep, appetite, and much else in the nervous system. None of that is in dispute. What was never nailed down, despite the metaphor's popularity, was the specific causal claim: that people with depression have measurably less of it, or less serotonin activity, than people without.

Joanna Moncrieff, working with Ruth Cooper, Tom Stockmann, Simone Amendola, Michael Hengartner, and Mark Horowitz, set out to check. They did not run a new experiment. They ran an umbrella review, a study of studies, gathering the major bodies of evidence that had accumulated on serotonin and depression and asking what they showed when read together rather than one at a time. Their review, published in Molecular Psychiatry in July 2022, drew on seventeen systematic reviews, meta-analyses, and large-dataset studies, searched up to a December 2020 cutoff. It covered five separate lines of evidence: serotonin and its breakdown product measured in blood and spinal fluid, the density of a specific serotonin receptor in the brain, the amount of a protein that recycles serotonin back into neurons, what happens to mood when serotonin's chemical building block is stripped from the diet, and whether the gene for that recycling protein is linked to depression risk.

Some of these sub-studies were large. A meta-analysis of the breakdown-product measurements covered 1,002 people. A cohort meta-analysis of blood serotonin covered 1,869. A gene-association study covered 115,257 people, and a separate collaborative meta-analysis of the same gene covered 43,165. Across all five lines, the review found no consistent association between any serotonin marker and depression. The receptor and recycling-protein findings that did turn up were weak and inconsistent, and could not be cleanly separated from the effect of prior antidepressant use itself — some evidence pointed the other way, toward long-term antidepressant use lowering serotonin concentration rather than depression doing so. The dietary-depletion studies, run on 566 healthy volunteers and 75 people with a family history of depression, showed no reliable mood effect either. Seventeen bodies of evidence, five different methods, and no wire connecting low serotonin to depression.

The review did not go unanswered. A group of thirty-six authors led by Sameer Jauhar and Philip Cowen, based at King's College London and Oxford, published a formal Comment in the same journal the following June, titled "A leaky umbrella has little value." Their argument was not that the umbrella review's raw numbers were wrong, but that its method was: they pointed to selective reporting, over-simplification, and interpretive errors, and focused their strongest objections on two specific literatures — the dietary-depletion studies and the brain-imaging studies of serotonin receptors — as the two areas they judged most directly capable of testing serotonin's role in depression at all. A related paper by Jauhar, Cowen, and Michael Browning, "Fifty years on: Serotonin and depression," revisited the original Coppen hypothesis across its full history and pushed back on the umbrella review's conclusions from that longer vantage point.

Moncrieff and her co-authors replied in the same journal. Their position, stated plainly, was that the burden of proof runs the other way: "the onus is on others to present a convincing case... not on us to disprove every speculation." That is the shape of the dispute as it stands. It is not a disagreement about whether any single number was miscounted. It is a disagreement about what absence of evidence is allowed to mean, and about which two out of five evidence streams should carry the most weight. Readers should know both papers exist, under whose names, and that neither side has retracted anything.

The Puzzle of the Second Week

A second thread of the popular story concerns timing. Antidepressants that raise serotonin availability do so within hours of the first dose — the pharmacology is fast. But patients and clinicians have long described the mood benefit as arriving only after two to six weeks, a gap the thermostat metaphor cannot explain on its own: if the chemical target is hit immediately, why does the room stay cold for a month? Matthew Taylor, Nick Freemantle, John Geddes, and Zubin Bhagwagar, at the University of Oxford, tested whether that "delayed onset" picture was itself accurate. They pooled 28 randomized controlled trials, 5,872 patients in total, split between 2,254 on placebo and 3,618 on an SSRI, and tracked how early measurable symptom improvement actually appeared.

They found it appeared early. By the end of the first week of treatment, patients on an SSRI were 1.64 times as likely as those on placebo to have already achieved a 50 percent reduction on a standard depression rating scale — a relative risk with a 95 percent confidence interval of 1.2 to 2.25, comfortably excluding no effect. A model in which benefit accumulates steadily from the first week, at a logarithmically decreasing rate, fit the pooled data better than a model of a true delayed switch flipping weeks later. Most of the total improvement still took several weeks to accrue, but the idea of a silent gap before anything happens does not survive the data. A later commentary by Barbui and Cipriani raised concerns about publication bias in the underlying trial literature — some negative trials may simply not have been published — but the early-onset finding itself has not been seriously overturned since.

None of this settles whether the drugs work. That is a separate, and better-answered, question, and it is worth pausing on why the two questions can have different answers at all. A drug can move a biological system in a way that helps a patient without that system being the one the popular story named, and without anyone yet knowing which system deserves the credit. Andrea Cipriani led an international collaboration through a network meta-analysis of 522 double-blind randomized trials, 116,477 participants, comparing 21 different antidepressants, published in The Lancet in 2018. Every one of the 21 drugs outperformed placebo on response rate at eight weeks, defined across the pooled trials as at least a 50 percent reduction on a standardized depression rating scale. The size of that advantage varied by drug, expressed as odds ratios against placebo ranging from 2.13 for amitriptyline down to 1.37 for reboxetine — real, but modest, and uneven across compounds. Some drugs that beat placebo, including fluoxetine and reboxetine, performed worse than other active drugs when compared head to head; others, including sertraline and escitalopram, were more tolerable, with lower dropout, expressed in the same analysis as tolerability odds ratios between 0.43 and 0.77 relative to less tolerable drugs.

A methodological critique published the same year in Frontiers in Psychiatry raised a fair caution about what "beats placebo" is allowed to mean: the trials mostly relied on clinician-administered rating scales rather than patients' own reports, and a statistically significant difference on such a scale is not automatically the same thing as a difference a patient would call meaningful in daily life. That caution has not displaced the standard reading of the Cipriani data — it has not shown the drug-placebo gap to be illusory — but it is a live objection about effect size and measurement, not settled either way, and it deserves to sit alongside the headline finding rather than be smoothed out of it.

Put the two halves next to each other and the honest picture has no single slogan. The chemical-imbalance story, read as a specific claim that depression is caused by low serotonin, has no support across seventeen bodies of evidence spanning five different methods — that is Moncrieff's finding, and it survives its critics' sharpest specific objections even if the critics dispute how much weight two of those five literatures deserve. At the same time, the drugs built on that story, whatever the reason they work, measurably outperform placebo for moderate-to-severe depression, with benefit visible within the first week and continuing to accrue for weeks after. The mechanism story was wrong, or at least badly oversold. The treatment effect, imperfect and debated in size, was not.

Overturned if: a large, pre-registered, methodologically uncontested study directly linking a serotonin marker to depression onset — of the kind absent from all seventeen reviews synthesized in 2022 — is published and independently replicated.

Sources

บทถัดไป · 2 / 1202. The Currency of Pursuit
Size
Page
Font

The Architecture of Affect

❦
101. The Myth of the Broken Thermostat202. The Currency of Pursuit303. The Cost of Chronic Vigilance404. The Architecture of Resilience505. Rapid Restoration606. Relaxing the Predictive Constraint707. The Mathematics of Disappointment808. The Logic of Giving Up909. Environmental Recalibration1010. The Distributed Self1111. The Uncharted Territory1212. Agency in a Predictive World
1 / 12Next ›