In July 2022, Joanna Moncrieff and colleagues published an umbrella review in Molecular Psychiatry that landed with unusual force: after decades of clinical dogma, they concluded there was no consistent evidence supporting the hypothesis that depression is caused by reduced serotonin activity or concentrations. The public reaction was extraordinary, not because the finding was novel to researchers, but because it exposed a gap between what the field quietly acknowledged and what patients had been told for a generation.

The chemical imbalance narrative was never a rigorous scientific theory. It was a heuristic—useful for reducing stigma, convenient for prescribing, and profoundly compatible with a pharmaceutical industry positioned to correct it. Yet the neurobiology of mood was always more complex than a single monoamine could accommodate.

This article revisits the evidentiary basis of the low-serotonin model, situates its persistence within the political economy of drug development, and considers what a more epistemically honest neuroscience of depression might look like. The goal is not to indict SSRIs, which demonstrably help many patients, but to interrogate how a mechanistic just-so story became indistinguishable from established fact—and what that conflation reveals about the epistemology of biological psychiatry.

Direct evidence for a serotonergic deficit in depression has always been remarkably thin. The classical strategies for testing the hypothesis included measuring 5-HIAA, serotonin's principal metabolite, in cerebrospinal fluid; assaying platelet serotonin transporter binding; and inducing acute tryptophan depletion in remitted and healthy subjects. None produced the clean, replicable signal that a straightforward deficiency model would predict.

CSF 5-HIAA studies yielded inconsistent findings across cohorts, with meta-analyses showing no reliable elevation or reduction in unmedicated depressed patients relative to controls. Receptor imaging using PET ligands such as [11C]WAY-100635 for 5-HT1A revealed heterogeneous alterations that varied with prior treatment history and clinical subtype, complicating causal inference.

Tryptophan depletion, arguably the most direct experimental manipulation, produced transient mood decrements in previously depressed patients but generally failed to induce clinical depression in healthy individuals with no personal or family history. This dissociation suggested vulnerability rather than causation—serotonin function was a modulator, not a lesion.

Genetic association studies compounded the difficulty. The serotonin transporter promoter polymorphism, once heralded as a robust moderator of depression risk, failed to replicate in large collaborative analyses. The candidate gene era yielded few durable findings, and contemporary GWAS results implicate polygenic architectures spanning synaptic, immune, and circadian pathways.

What the literature actually demonstrated was that serotonergic circuits are implicated in mood regulation without being deficient in any measurable, uniform way. The distinction matters. A neurotransmitter can be functionally relevant to a disorder without being causally depleted, just as insulin is relevant to glucose homeostasis in ways that extend well beyond simple insufficiency.

Takeaway

Mechanistic involvement is not the same as mechanistic deficit. Confusing the two collapses a nuanced regulatory system into a broken-part metaphor that neither the data nor the biology supports.

The serotonin hypothesis did not emerge from basic neuroscience and then find its therapeutic expression in SSRIs. The causal arrow ran the other way. In the 1960s, the observation that reserpine could induce depressive symptoms and that monoamine oxidase inhibitors alleviated them prompted Joseph Schildkraut's catecholamine hypothesis, later broadened to include serotonin after Alec Coppen's contributions.

The hypothesis was, from the start, a reverse inference: because drugs affecting monoamines altered mood, mood disorders must involve monoamine dysfunction. This is a logically fragile move. Aspirin relieves headache without headaches being caused by aspirin deficiency. Yet the pharmaceutical significance of the framing was immense.

When fluoxetine entered the market in 1988, its regulatory story dovetailed with a marketable neurobiological narrative. Direct-to-consumer advertising, permitted in the United States after 1997, then translated a contested hypothesis into a folk metaphor. Illustrations of synapses with too few dots, corrected by pills, populated magazine spreads and physician offices alike.

The academic literature was not immune. Industry-funded continuing medical education, sponsored symposia, and ghostwritten review articles amplified a coherent story that was easier to teach than the messy truth. Prescribers, in turn, communicated the model to patients because it destigmatised suffering by relocating it from character to chemistry.

The result was a rare epistemic phenomenon: a hypothesis sustained less by evidence than by its institutional utility. It served patients seeking legitimacy, clinicians seeking parsimony, and companies seeking markets. That alignment of incentives made critical scrutiny professionally awkward and slowed the field's willingness to publicly abandon a framework its own leading researchers privately considered inadequate.

Takeaway

When a scientific model becomes commercially and clinically load-bearing, its evidentiary weaknesses become socially expensive to name. Ask who benefits from a theory's persistence, not only whether it fits the data.

Discarding the deficiency model does not require discarding serotonin. Contemporary work frames 5-HT signalling in terms of neural plasticity, predictive processing, and network dynamics rather than tonic levels. The Carhart-Harris and Nutt REBUS framework, for instance, positions 5-HT2A activation as a modulator of the precision-weighting of prior beliefs, offering a computational account that explains both SSRI and psychedelic effects without invoking depletion.

Emerging evidence also implicates neurogenesis in the dentate gyrus, BDNF signalling, glutamatergic modulation via NMDA and AMPA receptors, and glial-immune interactions as convergent substrates of antidepressant response. Ketamine's rapid efficacy, largely orthogonal to serotonergic mechanisms, has forced the field to accept therapeutic pluralism at the mechanistic level.

This pluralism argues for what might be called mechanism-agnostic evaluation. If the pathway to clinical improvement is not reducible to a single neurochemical correction, then treatment trials should be judged by outcome heterogeneity, durability, and functional recovery rather than by their alignment with a preferred molecular story.

Predictive markers—inflammatory profiles, sleep architecture, reward learning parameters, resting-state connectivity—may eventually stratify patients more usefully than diagnostic categories built on symptom lists. Trials such as EMBARC have begun this stratification, though replication and generalisability remain open questions.

The broader lesson concerns epistemic humility. Psychiatry's periodic infatuations with single-mechanism explanations—dopamine for schizophrenia, cortisol for depression, gut microbiota for nearly everything—reflect a discipline seeking the reductive clarity of its neighbouring fields. Complex phenotypes will likely require probabilistic, multi-scale models. That is a slower, less marketable story, but a more defensible one.

Takeaway

Effective treatment does not require a correct mechanism, and a correct mechanism will rarely be singular. Progress lies in tolerating explanatory pluralism rather than resolving it prematurely.

The serotonin hypothesis is a case study in how scientific claims acquire authority disproportionate to their evidentiary base. It persisted because it was useful, teachable, and commercially reinforced, not because the data compelled it.

Its unraveling need not destabilise clinical practice. SSRIs remain reasonable options for many patients, and their empirical track record stands independent of the mechanistic story once told to justify them. What must change is the epistemic framing around them.

For researchers, the task ahead is to build models that respect the layered, dynamical nature of mood pathology—models that can be wrong in specific, testable ways. For clinicians, it is to communicate uncertainty without eroding therapeutic hope. That balance, though difficult, is where a more honest psychiatry begins.