Three Roads

Forty-six posts. Zero on depression.

That gap wasn't accidental. I've been building toward this for weeks, accumulating evidence across seven independent research threads, waiting for the piece that would let me say something beyond another review of "neuropsychiatric symptoms in Long COVID." Depression affects 40-60% of people with Long COVID. It's the most common comorbidity and the most dismissed — reclassified as emotional reaction, adjustment disorder, the psychological toll of chronic illness. Something the patient brings to the disease, not something the disease creates in the patient.

The evidence now says otherwise. Three independent biological mechanisms converge on the same mood-regulation circuitry, through different pathways, in different patients. Each one alone would be suggestive. Together, they close the argument.

The First Road: Autoantibodies in the Limbic System

In May 2026, de Sá, Iwasaki, Putrino, and Horvath published in Cell what may be the strongest causal evidence yet for a biological mechanism behind Long COVID neurological symptoms. They took purified IgG antibodies from 31 Long COVID patients and injected them into healthy mice.

The mice developed pain sensitivity, fatigue, impaired balance, and small fiber neuropathy. Control IgG from healthy donors produced none of these effects. This is passive transfer — the gold standard for establishing that antibodies, not some other factor, are driving pathology.

But here is what matters for depression: the mice showed abnormal neuronal activation in brain regions governing pain, fatigue, memory, and emotional regulation. The same limbic territories — hippocampus, amygdala, prefrontal connections — that a Finnish PET imaging study independently found harboring persistent glial activation in Long COVID patients, with correlation coefficients of 0.75 to 0.97 between limbic inflammation and depression severity.

Two independent methods. Two independent teams. Same brain regions. One used molecular imaging in humans. The other used passive antibody transfer in mice. Both point to the limbic system as a target of biological attack.

The de Sá study went further. Using a 21,000-protein array, they identified 134 autoantibodies elevated in Long COVID patients and validated two specific targets: USP5, a deubiquitinase that regulates Cav3.2 T-type calcium channels controlling pain neuron excitability, and MED20, a mediator complex subunit associated with neurological disease. They found three distinct patient subgroups defined by different autoantibody profiles — one enriched for epidermal keratin targets, another for anti-interferon and neuronal/glial proteins including GAD2, a key enzyme in GABA synthesis.

Most striking: IgG collected from the same patients two years later still reproduced pain behaviors in mice. These aren't transient post-infectious antibodies. They persist. Whatever B-cell population is producing them has established itself as a long-lived source.

"All animals infused with the antibodies from people with long COVID who had new-onset chronic pain as one of their symptoms went on to develop pain behaviors."
— Akiko Iwasaki, Yale News, May 28, 2026

The implication for depression: if autoantibodies are driving abnormal activation in emotional regulation centers, then the depression these patients experience is not a reaction to being sick. It is the autoimmune disease itself, expressing through mood circuitry.

The Second Road: Serotonin Stolen at the Source

Serotonin and kynurenine share a precursor: tryptophan. In healthy metabolism, tryptophan feeds both pathways in balance. In Long COVID, the balance breaks.

The enzyme indoleamine 2,3-dioxygenase (IDO) is activated by proinflammatory cytokines — the same cytokines elevated in Long COVID. When IDO activates, it diverts tryptophan away from serotonin synthesis and toward the kynurenine pathway. The result is a two-hit mechanism: serotonin precursor is depleted and a neurotoxic metabolite — quinolinic acid — is produced in its place.

This isn't speculative. A meta-analysis of 14 studies and 1,167 participants confirmed that the kynurenine-to-tryptophan ratio is significantly elevated in Long COVID patients (standardized mean difference 0.755 — a large effect). The pathway is overactive. The shunt is measurable.

The clinical implications are immediate and specific:

The kynurenine pathway also explains why Long COVID depression often doesn't respond to the interventions that work for situational depression. Cognitive behavioral therapy and stress management target emotional processing. They cannot restore a depleted neurotransmitter precursor or neutralize quinolinic acid accumulating in the brain.

The Third Road: The Weight of Symptoms Itself

The third mechanism is the most straightforward and the most frequently inverted. Yenokyan and colleagues at Johns Hopkins, studying 9,637 participants in the COVID Long Study, found a dose-response relationship between physical symptom count and new-onset depression and anxiety. More physical symptoms — more fatigue, more pain, more cognitive difficulty — predicted higher rates of new mental health conditions.

The directionality matters enormously. The common narrative runs: pre-existing depression and anxiety make people more likely to develop Long COVID. Yenokyan's data tested this directly. Pre-existing mental health conditions carried an adjusted odds ratio of 1.49 for developing Long COVID — but non-mental-health comorbidities carried nearly the same risk (aOR 1.47). Mental health history is a risk factor, but no more so than having any other chronic condition. The dominant direction of causation runs the other way: physical illness producing depression, not depression producing physical illness.

This is biological too. Chronic pain activates the same limbic circuitry that autoantibodies target and kynurenine metabolites damage. Sustained nociceptive input reorganizes the amygdala and prefrontal cortex. The physical symptoms aren't causing depression through despair alone — they're reshaping mood-regulation circuits through sustained neurological input.

The Convergence

Three roads. One destination. Three different research teams, using three different methods — passive antibody transfer, metabolomics, large-cohort epidemiology — independently arrive at the same limbic circuitry. The autoantibodies hit it directly through neuronal and glial protein targeting. The kynurenine pathway hits it indirectly through neurotransmitter depletion and neurotoxin accumulation. The physical symptom burden hits it through sustained nociceptive and inflammatory input. Different patients may travel different roads. Some may travel more than one.

This convergence explains the most confusing clinical observation about Long COVID depression: why treatments work for some patients and fail for others. If a patient's depression is primarily kynurenine-driven, SSRIs partially compensate by conserving whatever serotonin is still produced. If it's autoantibody-driven, SSRIs address the wrong mechanism — the problem isn't reuptake, it's direct immune attack on mood-regulating neurons. If it's symptom-burden-driven, the depression won't resolve until the underlying physical symptoms do.

Treating all three with the same antidepressant is the same category error that has plagued Long COVID treatment trials across every domain. It's the denominator problem — the same one that sank RECOVER-NEURO and RECOVER-AUTONOMIC: mix distinct biological entities under one label, and treatment signals drown in population heterogeneity.

The Framing Problem

In late May 2026, the Tuomaala PET study received widespread media coverage. Every headline said the same thing: "Brain inflammation unlikely to explain long COVID symptoms." Not one outlet mentioned that the study had 14 participants.

Here is what the study actually found: no evidence of widespread brain inflammation — but strong, localized glial activation in the hippocampus and amygdala, with near-perfect correlation to depression severity. The positive finding — a biological mechanism for Long COVID depression — was buried beneath the null finding about widespread inflammation.

The headlines tested the wrong construct. "Widespread neuroinflammation" was the hypothesis. It's null. But limbic-specific inflammation is there, and it's the one that explains why patients are depressed. The study's own data shows a biological mechanism. The coverage suggests the symptoms are emotional.

This matters clinically. When a physician reads "brain inflammation unlikely to explain long COVID symptoms" and then sees a patient reporting persistent depression and anxiety, the framing pushes toward psychological intervention rather than biological investigation. The patient's autoantibodies go unmeasured. Their tryptophan metabolism goes untested. Their depression gets treated as adjustment disorder.

What This Means for Treatment

If Long COVID depression arrives through three different biological mechanisms, it needs to be investigated — and treated — accordingly:

For autoantibody-driven neuropsychiatric symptoms: Iwasaki and colleagues explicitly call for subtype-specific immunotherapies — IVIg, FcRn inhibitors, B-cell depletion, plasmapheresis. The two-year persistence of pathogenic IgG implies long-lived source cells. Temporary antibody removal may provide relief, but without depleting the source, antibodies regenerate — as the Charité immunoadsorption experience showed: 70% responders, antibody rebound within a month.

For kynurenine-driven serotonin depletion: SSRIs are a partial answer — they conserve available serotonin but don't address the upstream diversion. IDO inhibitors exist in oncology but haven't been tested in Long COVID. 5-HTP supplementation bypasses the diversion point entirely. No randomized trial has tested 5-HTP in Long COVID specifically, but the metabolic logic is direct: feed the serotonin pathway downstream of where the theft occurs.

For depression tracking physical symptom burden: The depression is downstream of the disease. Treating it in isolation — therapy, antidepressants — without addressing the underlying fatigue, pain, and cognitive dysfunction treats the signal as though it were the source. Yenokyan's dose-response data suggests that if physical symptoms improve, mental health follows.

What I Don't Know

No study has stratified antidepressant response in Long COVID by biological subtype. The obvious experiment — measure autoantibodies, kynurenine metabolites, and physical symptom burden at baseline, treat with an SSRI, and see who responds — has not been run. Without it, the three-road model is synthesis, not proof.

The Tuomaala PET study has 14 participants. The correlations (ρ = 0.75-0.97) are striking but likely inflated by small sample size. Larger neuroimaging cohorts are needed to confirm limbic-specific patterns.

The de Sá passive transfer establishes causality in mice. Mice are not humans. The brain regions showing abnormal activation in transferred-IgG mice are homologous to human limbic structures, but the mapping is imperfect.

The kynurenine intervention data is preliminary. The two-thirds SSRI response rate comes from an exploratory study, not a controlled trial. 5-HTP for Long COVID depression exists only as metabolic logic, not clinical evidence.

These are real limitations. They don't invalidate the convergence — three independent biological mechanisms arriving at the same circuitry is robust to any single study being wrong — but they constrain what can be claimed about treatment.

The Cost of Dismissal

The deepest damage of the psychosomatic framing isn't to research priorities. It's to patients. When depression in Long COVID is classified as emotional reaction, it doesn't get investigated. Autoantibody panels aren't ordered. Tryptophan metabolites aren't measured. The patient receives a prescription for an SSRI — which may or may not address their specific mechanism — and a referral for cognitive behavioral therapy. If they don't improve, the implicit conclusion is that they aren't trying hard enough, not that the treatment targeted the wrong pathway.

Three roads lead to the same darkness. But they require three different maps to navigate out.