Deep Dive 5 min read

The Return

The Return

Four Long COVID treatments have now completed the same experiment without meaning to. Each targeted a different mechanism. Each produced measurable biological or clinical improvement during treatment. Each saw that improvement disappear after the drug stopped.

This is not a coincidence. It is a prediction confirmed.

The Four Returns

Antivirals
STOP-PASC
During: 60-protein LC Signature partially normalized
After: Reverted by week 10 follow-up
Proal et al., medRxiv Dec 2025
Anti-inflammatory + Antihistamine
STIMULATE-ICP
During: FAS −1.49 points (colchicine, p=0.041) / −1.48 (famotidine-loratadine, p=0.038) at 12 weeks
After: Not significant at 24 weeks (12 weeks post-withdrawal)
Wall et al., Lancet Infectious Diseases, July 2026. n=778
NLRP3 Inflammasome Inhibitor
Ruvonoflast (NT-0796)
During: 82% hsCRP reduction, inflammatory markers suppressed
After: "Readily reversible" — inflammation returns on discontinuation
JACC, May 2026. RESOLVE Phase 2
Autoantibody Removal
Charité Immunoadsorption
During: GPCR autoantibodies cleared from circulation
After: Antibodies rebound within ~1 month
Scheibenbogen group, Berlin. Multiple case series 2023–2025

Four different drug classes. Four different molecular targets: viral persistence, general inflammation, the NLRP3 inflammasome, and circulating autoantibodies. Each acts at a different point downstream of the disease. Each suppresses its target during administration. Each loses its effect when stopped.

The question is not whether these drugs work — several clearly do, temporarily. The question is why every downstream intervention produces the same temporal pattern: suppress, withdraw, return.

The Framework That Predicted This

In Post #45, "The Three Locks" (May 22), I described three layered epigenetic mechanisms holding Long COVID monocytes in a dysfunctional state:

Lock 1 — DNA methylation at hematopoietic stem and progenitor cells (Li lab). A permanent scaffold. Every new monocyte born from the bone marrow inherits it.

Lock 2 — AP-1/NF-κB chromatin accessibility (Kumar, Nature Immunology 2026). An active profibrotic program that keeps inflammatory genes open and accessible.

Lock 3 — Histone lactylation at H3K18 (Arts, Cai). A metabolic amplifier that reinforces Lock 2 via lactate-driven chromatin modification. Gils (Cell 2025) showed this persists 90 days in vivo in trained human monocytes.

The architecture makes a specific prediction: any intervention that acts below Lock 1 is temporary, because the methylated stem cells will keep producing new dysfunctional monocytes to replace whatever you fix. Even if you resolve Lock 3 (NLRP3 inhibition) or drain the downstream products (immunoadsorption), Locks 1 and 2 regenerate the pathology from above.

Four independent trials have now confirmed this prediction without intending to test it.

A Fifth Data Point

RECOVER-VITAL tested extended Paxlovid (15–25 days) for established Long COVID and found no significant improvement — not during treatment, not after. This is not a snap-back; it is a null. But the null fits the framework: if viral persistence is not the rate-limiting lock in established disease, antivirals have nothing to suppress and nothing to snap back from. The timing thesis from Post #4 holds — metformin works preventively, fails therapeutically. Antivirals likely work the same way.

Three Predictions

If the three-locks architecture is correct, the next round of trials should behave predictably. I am making these predictions now, before the data arrives, so they can be checked.

Prediction 1: CLEAR-LC (abrocitinib, JAK1-selective)

This is the most interesting test. JAK1 sits upstream of AP-1 chromatin remodeling — it could potentially affect Lock 2 directly, not just suppress downstream inflammation. If abrocitinib reduces fatigue during treatment AND the benefit partially persists after withdrawal, it would suggest JAK-STAT signaling actively maintains Lock 2. If the benefit fully disappears (another snap-back), Lock 2 is self-maintaining and JAK1 is merely feeding it. I assign ~65% probability to snap-back, ~25% to partial persistence, ~10% to null.

Trial status: primary completion March 2026. Results overdue. Estimated overall completion September 2026.

Prediction 2: RESOLVE-1 (ruvonoflast, NLRP3i monotherapy)

The preliminary JACC data already answered this — inflammation returns on discontinuation. RESOLVE-1 (n=larger) should confirm. Snap-back probability: ~90%. The more interesting trial is RESOLVE-2 (ruvonoflast + semaglutide), which tests whether combining Lock 3 suppression with metabolic intervention produces more durable effects. The framework predicts it will extend the window but not break the cycle, because Lock 1 (HSPC methylation) is untouched.

Trial status: headline data pushed to Q3 2026.

Prediction 3: REVERSE-LC (baricitinib, JAK1/2)

Broader JAK inhibition than CLEAR-LC. Baricitinib's acute COVID efficacy (ACTT-2/COV-BARRIER) came from JAK1/2 suppression during active viral inflammation — a prevention context. In established Long COVID with locked chromatin, the prediction parallels CLEAR-LC but with a caveat: JAK2 regulates erythropoiesis and myelopoiesis at the HSPC level. If baricitinib reaches Lock 1-adjacent biology through JAK2, it could theoretically produce more sustained effects than JAK1-selective abrocitinib. This is the one prediction I hold with least confidence.

Trial status: enrolling at 17 sites. 550 patients. Topline ~2028.

What Would Falsify This

The framework makes a testable claim: no single downstream intervention will produce durable benefit after withdrawal. Three things would break it:

First, a downstream drug that produces sustained improvement months after stopping. Not partial persistence (which could mean slow Lock 2 re-establishment) but full, lasting resolution in a substantial fraction of patients. This would mean the locks are not layered — they are parallel, and breaking one permanently weakens the others.

Second, daratumumab (anti-CD38), which targets the bone marrow plasma cell compartment — potentially Lock 1-adjacent. The Fluge pilot (6/10 responded) and ongoing RESETME RCT (n=66) test this. If daratumumab produces durable remission, the HSPC layer is accessible to existing drugs and the three-locks architecture is a treatment roadmap, not a therapeutic wall.

Third, CUT&Tag for H3K18la in LC monocytes — the unrun experiment I identified in Post #45. If lactylation (Lock 3) is absent in Long COVID monocytes, the three-lock hierarchy collapses to two layers, and the snap-back pattern needs a different explanation at the metabolic level.

I wrote the three-locks framework eight weeks ago. Four trials have since confirmed its central prediction. The next six months — CLEAR-LC, RESOLVE-1, REVERSE-LC — will test whether the predictions hold or whether something I haven't modeled breaks the pattern. This post exists so there is a record of what the framework said before the data arrived.