For most of the past thirty years, the story about depression that reached the public was tidy: it is a chemical imbalance, specifically a deficit of serotonin, and antidepressants correct it. That story is not exactly wrong, but it is not quite right either, and the gap between what researchers actually believe and what patients are often told has become one of the more consequential misunderstandings in modern medicine.
As of 2026, the science of depression is both more sophisticated and more humble than the chemical imbalance narrative allowed. Here is a careful look at where the evidence stands.
What we know
Depression is real, biological, and heterogeneous. Twin studies consistently show meaningful heritability. Neuroimaging finds reliable, if modest, differences in activity and connectivity in regions like the subgenual anterior cingulate and the default mode network. Inflammation appears to play a role in a subset of cases. Sleep architecture is often disrupted before mood symptoms appear.
We also know that depression responds to treatment. The APA and NIMH both recognize several first-line options with substantial evidence bases:
- Cognitive behavioral therapy and interpersonal therapy produce meaningful improvement in a majority of patients with mild-to-moderate depression.
- SSRIs, SNRIs, and atypical antidepressants work better than placebo, particularly for moderate-to-severe cases, though effect sizes are more modest than early marketing suggested.
- Exercise, especially aerobic exercise at moderate intensity, has antidepressant effects comparable to some medications for mild-to-moderate cases.
- Repetitive transcranial magnetic stimulation (rTMS) and, for treatment-resistant cases, ketamine derivatives have expanded the toolkit considerably in the past decade.
What we do not know
The honest answer to "what causes depression" is: we do not fully know. The serotonin hypothesis, in its simple form, has not held up. A widely discussed 2022 umbrella review found no consistent evidence that low serotonin levels cause depression, which does not mean serotonergic medications do not work, only that the mechanism by which they help is more complicated than a deficit-and-refill model.
The chemical imbalance story was useful shorthand. It was not a mechanism.
We also do not know why any given antidepressant works for one person and not another, why some episodes remit spontaneously, or why some people experience a single episode and others cycle through many. Predictive biomarkers remain elusive despite substantial investment.
The heterogeneity problem
One reason the science is messy is that "depression," as clinically defined, almost certainly encompasses several different biological entities that happen to produce similar symptoms. A person whose depression follows chronic inflammation may need different treatment than someone whose depression follows sleep deprivation, grief, or hormonal transition. The DSM-5-TR captures this only crudely.
Researchers are increasingly moving toward transdiagnostic and dimensional frameworks, like NIMH's Research Domain Criteria, that try to identify underlying processes across diagnoses rather than treating depression as a single thing.
What this means for patients
If you have depression, the practical implications are less dramatic than the scientific shift might suggest:
- Treatment still works. Most people who engage in evidence-based care improve.
- The first thing you try may not be the thing that helps. Expect iteration.
- Both therapy and medication have evidence behind them. Combined treatment often outperforms either alone for moderate-to-severe cases.
- Lifestyle factors are not a substitute for treatment, but they are not decorative either. Sleep, movement, and social connection have measurable effects.
The bottom line
We understand depression better than we did a generation ago, and we are more honest about what we still do not understand. The chemical imbalance shorthand is fading, but the treatments built on that era of research continue to help people. What is changing is the promise of a single explanation, replaced by something more useful: a growing set of tools matched, imperfectly but improvingly, to a condition that turns out to be many conditions.