Depression is not one thing, it has phenotypes

The data on subtypes has existed for years. Immunometabolic depression (IMD), treatment-resistant depression with high versus low inflammatory profiles, cyclic depression, reactive depression.
Each subtype carries different underlying biology and responds to different interventions. Yet in most clinical environments, subtyping remains binary: responsive versus treatment-resistant.
That framing is not neutral. It almost locates the failure in the patients biology rather than in the match between biology and treatment.

My master thesis was titled "Comparing Pro-Resolution and Anti-Inflammatory Therapies for Immune Dysregulation in Depression: Mechanisms and Clinical Efficacy."
That work made the subtyping problem impossible to unsee.

A paper in Brain, Behavior and Immunity makes the same argument with hard data.
https://pubmed.ncbi.nlm.nih.gov/39477079/

Around 30% of people with major depression carry a distinct biological signature: elevated inflammatory markers (CRP, IL-6, TNF-α), metabolic syndrome dysregulation, and atypical energy-related symptoms including hypersomnia, increased appetite, fatigue and leaden paralysis. This cluster is called immunometabolic depression.
The MOTAR study (n=141, 16 weeks) compared running therapy (3x/week, 70-85% heart rate reserve) against SSRIs in patients with depression and anxiety. On general depressive symptom scores, both interventions produced comparable reductions. Remission rates were 45% for running therapy and 42% for antidepressants. Neither treatment was superior for mood on the whole group.

But on the immunometabolic dimension, the divergence was strong.
Running therapy decreased the composite IMD index significantly (Cohen's d = 0.85, p < 0.001). The antidepressant group moved in the opposite direction: the IMD index increased. Metabolic syndrome markers, lipid profiles, and inflammatory indices all trended worse with SSRIs, consistent with their known effects on appetite, weight and glucose-insulin homeostasis.

This is the important nuance: exercise did not produce better mood scores overall (within the study time frame), but it produced substantially better biological outcomes in the exact domain where this subtype is dysregulated. The two treatments are not interchangeable. They hit different targets.

A cleaner immunometabolic profile is a physiological invitation for better mood whereas a pro-inflammatory one is not.

When a patient with IMD receives an SSRI and does not respond, the clinical label becomes treatment-resistant depression. But if the intervention never addressed the underlying immunometabolic dysregulation, resistance is the wrong word. It is a mismatch.

Exercise is not a soft lifestyle recommendation.
For some subtypes it may be the primary intervention.