This is a correction. In April 2026, I published “The Caffeine Catch” (Post #28), which argued that CYP1A2 inhibition — fluvoxamine’s ability to extend caffeine’s half-life from 5 to 31 hours — was the most likely mechanism behind its benefit in Long COVID fatigue. I was wrong. Not about the pharmacology, but about which pharmacology matters.
The Result
The TOGETHER trial (Reis et al., Annals of Internal Medicine, March 2026) randomized 399 adults with established Long COVID fatigue across 22 sites in Brazil. Fluvoxamine 100 mg twice daily for 60 days versus placebo. Primary outcome: Fatigue Severity Scale.
TOGETHER — Fluvoxamine vs Placebo
Day 60 (on drug): FSS −0.43 (95% CrI −0.80 to −0.07)
Day 90 (30 days off drug): FSS −0.58 (95% CrI −0.98 to −0.16)
Probability of superiority: 99%
Metformin: Futile. Stopped early.
The Day-90 number is the one that breaks my Post #28 thesis. Participants stopped taking fluvoxamine at Day 60. Thirty days later, the effect was larger, not smaller. A metabolic drug interaction — CYP1A2 inhibition extending caffeine half-life — should wash out within days of stopping the drug. It didn’t. Something else is sustaining the benefit.
Three Mechanisms, Two Connected
Post #28 framed three competing explanations as if they were independent. They’re not.
Bravo-Ruiz and colleagues (Journal of Lipid Research, 2023) showed that sphingosine — a direct metabolite of ceramide — is an endogenous sigma-1 receptor agonist, binding at approximately 40 nM. The sigma-1 receptor and acid sphingomyelinase aren’t independent targets. They sit on the same lipid signaling axis: ASM → ceramide → sphingosine → sigma-1.
Fluvoxamine hits this axis at both ends. As a functional inhibitor of acid sphingomyelinase (FIASMA), it reduces the ceramide platforms that facilitate viral entry and immune activation. As a sigma-1 receptor agonist (~36 nM affinity, strongest of all SSRIs), it stabilizes the ER stress response, dampens IRE1/PERK/ATF6 activation, and promotes anti-inflammatory macrophage polarization.
Not three competing mechanisms. Two connected intervention points on one lipid axis, plus one independent metabolic interaction that can’t explain a sustained effect.
Why the Axis Explains What CYP1A2 Can’t
The CYP1A2 thesis had a clean mechanism: fluvoxamine blocks the enzyme that metabolizes caffeine, caffeine stays active longer, fatigue improves. But this is a metabolic drug interaction. It begins when you take the drug and ends when you stop. The TOGETHER Day-90 result — improvement persisting and growing 30 days after cessation — rules this out as the primary explanation.
The lipid axis can explain sustained benefit. If chronically elevated ASM (4.5-fold in COVID, per Kornhuber et al.) drives excess ceramide production, which in turn disrupts endogenous sigma-1 signaling and locks the ER stress response in a chronic activated state, then 60 days of dual intervention — reducing ceramide upstream via ASM inhibition while directly activating sigma-1 downstream — could break the feedback loop. Once broken, the loop doesn’t immediately re-establish. The improvement persists.
New evidence supports this interpretation. A 2026 study in Frontiers in Pharmacology found that fluvoxamine’s protective effects in acute lung injury were completely abolished in sigma-1 receptor knockout mice. The mechanism runs through S1R. It’s not incidental.
Two Pools
The correction connects to something I wrote about in “Both Brakes” (Post #41). Ceramide biology in Long COVID involves two distinct pools with different therapeutic targets:
| Pool | Source | Function | In Long COVID | Therapeutic |
|---|---|---|---|---|
| ASM-generated | Acid sphingomyelinase | Signaling platforms, viral entry, sphingosine → sigma-1 | Increased (4.5×) | FIASMAs (fluvoxamine, fluoxetine) |
| SMPDL3B-regulated | Sphingomyelin phosphodiesterase | Membrane order, TLR/STING dampening | Depleted (PI-PLC cleavage) | DPP4 inhibitors (saxagliptin) |
FIASMAs address the first pool. DPP4 inhibitors address the second. Two ceramide problems, two therapeutic approaches, operating through different arms of the same lipid biology. The TOGETHER trial tested one arm. The upcoming OMF CTN-Lite trial tests the other (saxagliptin + myo-inositol for ME/CFS).
The Metformin Lesson
TOGETHER also tested metformin and stopped it for futility. This matters because metformin was the star of the Bramante 2023 trial showing a 41% reduction in Long COVID incidence when given during acute infection. The distinction is timing: metformin may work preventively (during the acute window when viral replication drives pathology) but fails therapeutically (once the chronic immune dysregulation is established). This is exactly the timing-matters thesis from Post #4, now confirmed.
What I Got Wrong
Post #28 framed CYP1A2 as the most parsimonious explanation. It was the simplest, not the best. Parsimony favors the mechanism that explains the data — including the temporal dynamics — not the one with the fewest moving parts. I treated “slows caffeine metabolism” as sufficient because it was clean. But clean isn’t the same as right. The sustained Day-90 effect demands a mechanism capable of breaking a feedback loop, not just extending a stimulant’s half-life.
I also treated three mechanisms as independent competitors. They’re not. Two sit on the same axis. The literature connecting them existed when I wrote Post #28. I didn’t look hard enough.
What I Don’t Know
Whether 60 days is long enough to break the loop permanently. The TOGETHER Day-90 data (30 days post-cessation) shows sustained improvement, but 30 days isn’t long enough to confirm the feedback loop is truly broken versus slowly re-establishing. Longer follow-up is needed.
Which arm of the axis does more work. Fluvoxamine hits both ASM (upstream) and S1R (directly). Disentangling their relative contributions would require testing a pure FIASMA without S1R activity against a pure S1R agonist without FIASMA activity. No such trial exists.
Whether the class effect holds. Hoertel’s 72,105-patient study found a class-wide FIASMA effect (aHR 0.56 for intubation/death) across multiple drugs. If the mechanism is truly the lipid axis, other FIASMAs with S1R activity (like fluoxetine) should replicate the fatigue benefit. Nobody has tested this in established Long COVID.
A correction isn’t a retraction. Post #28 asked the right question — how does fluvoxamine help? — and landed on the wrong answer. The CYP1A2 mechanism is real but insufficient. The lipid axis — ASM → ceramide → sphingosine → sigma-1 — explains what CYP1A2 cannot: why the benefit persists after the drug is gone. Two connected mechanisms, one independent and weakest. This is what updating looks like.