In March 2026, researchers at Amsterdam UMC and UMC Utrecht published a striking result: take IgG antibodies from Long COVID patients, inject them into healthy mice, and the animals develop the disease. Pain hypersensitivity. Reduced locomotion. Persistent symptoms lasting weeks. The antibodies carried a blueprint of illness.
Six months earlier, a team in Barcelona published an equally striking result: take those same antibodies out of Long COVID patients via therapeutic plasma exchange, achieve a 70% reduction in circulating IgG, and… nothing happens. No improvement in function. No improvement in symptoms. No improvement in cognition. Not at day 8, not at day 15, not at day 90.
Transfer the disease with antibodies: it works. Remove the antibodies to cure the disease: it fails.
This is the autoantibody paradox. And its resolution tells us something fundamental about where Long COVID actually lives.
The Proof: Disease in a Syringe
Chen et al. didn't just show that antibodies cause harm. They showed that different antibodies cause different harm. From 34 Long COVID patients at Amsterdam UMC, the team isolated IgG and stratified it by biomarker profile into three subgroups:
The most unsettling finding: when they collected IgG from the same patients two years later, the antibodies reproduced identical symptom patterns in new mice. The blueprint doesn't fade.
This wasn't the first passive transfer experiment to succeed. In 2021, Goebel et al. showed that IgG from fibromyalgia patients transferred pain and nerve damage to mice — antibodies binding satellite glial cells, rewiring pain signaling. The mechanism is real across multiple chronic pain conditions. Autoantibodies aren't bystanders. They are functional drivers of specific symptoms.
The Failure: 70% Reduction, Zero Effect
If autoantibodies drive symptoms, removing them should help. España-Cueto et al. tested exactly this in a Phase II, double-blind, placebo-controlled trial at Hospital Clínic de Barcelona.
50 patients. Six sessions of therapeutic plasma exchange. 70% IgG reduction achieved. 40% were ANA-positive at baseline — four of ten became ANA-negative after treatment.
Result: no improvement on any endpoint, at any timepoint, in any subdimension.
Not function. Not symptomology. Not quality of life. Not cognition. Not biochemistry. Not at day 8, 15, 22, 45, or 90. The antibodies were measurably reduced. The disease was unmoved.
The Barcelona team even documented that some patients' autoantibodies shifted — ANA-positive patients becoming negative, titers dropping. The immune markers changed. The patient didn't.
Why the Paradox Isn't Actually Paradoxical
The transfer experiment and the treatment trial aren't contradictory. They're revealing complementary truths about the architecture of the disease.
When Chen's team injected IgG into mice, they introduced a finished product — a concentrated dose of pathogenic antibodies with no source to replenish them once the mice cleared them. The mice developed symptoms for weeks, then recovered. The experiment proved that autoantibodies can cause Long COVID symptoms. It didn't prove they're the root cause.
When España-Cueto's team drained IgG from patients, they removed the flood — but left the factory running. Within days, the bone marrow's long-lived plasma cells began replenishing exactly what was taken.
A Yale preprint from March 2026 adds another dimension. Chakravar et al. screened cerebrospinal fluid from 31 neurological Long COVID patients and serum from 73 post-COVID cognitive cases. Their phage immunoprecipitation sequencing found no shared autoantibody signature across cohorts. The diagnostic AUC was 0.62 — barely better than a coin flip. There is no single autoantibody target. The disease is heterogeneous at every level.
This means blanket removal — draining all IgG indiscriminately — is fighting a heterogeneous enemy with a uniform weapon. Even if removal worked, you'd need to know which antibodies to target in which patient. And you'd need to stop the source.
The Source
The source is the bone marrow.
Readers of this series will recognize the thread. In Post #19, I traced how SARS-CoV-2 epigenetically reprograms hematopoietic stem and progenitor cells (HSPCs), producing monocytes locked into an inflammatory state. In Post #20, I showed how CD38 — a surface enzyme on these dysfunctional cells — depletes NAD+, creating a metabolic trap that propagates exhaustion from cell to cell.
Now the autoantibody story connects to the same machinery. The bone marrow doesn't just produce broken monocytes. It harbors long-lived plasma cells (LLPCs) — antibody factories that can persist for years or decades.
LLPCs are immunology's cockroaches:
- Non-dividing — cyclophosphamide and other anti-proliferative drugs can't touch them
- Radiation-resistant — they survive doses that kill active immune cells
- Self-sustaining — they don't need B cell precursors or ongoing antigen to maintain production
- CD20-negative — rituximab (B cell depletion) can't reach them
- Deep in the marrow — they sit in survival niches that shield them from most drugs
Plasma exchange drains the pool. LLPCs refill it. The math never works out.
Three Tiers of Intervention
Science is slowly learning where this disease actually lives. The progression from crude to targeted tells the story:
Tier 1: Drain Everything (Failed)
Therapeutic plasma exchange is a blunt instrument. It removes all circulating proteins — pathogenic autoantibodies, protective antibodies, albumin, clotting factors — everything. The España-Cueto trial showed this doesn't work. Some uncontrolled studies from the Berlin Charité showed modest signals with immunoadsorption (79% IgG reduction, some symptom improvement), but without a control group, natural fluctuation can't be ruled out. The 14% spontaneous recovery rate in the Barcelona screening population underscores this problem.
Tier 2: Filter Selectively (Testing)
Efgartigimod (Vyvgart, argenx) represents a more sophisticated approach. This humanized IgG Fc fragment blocks the neonatal Fc receptor (FcRn), which normally rescues IgG from lysosomal degradation. Block FcRn, and IgG gets shunted to lysosomes for destruction — a targeted depletion that leaves IgM, IgA, and albumin untouched. FDA-approved for myasthenia gravis since 2021, it now has a registered trial for post-COVID POTS (~42 patients, 24 weeks).
It's smarter than plasma exchange. But it still only targets circulating antibodies, not the cells making them. If LLPCs are continuously producing pathogenic IgG, efgartigimod would need to be given indefinitely — treating the flood, never reaching the factory.
Tier 3: Kill the Factory (Promising)
This is where the thread from Post #20 reconnects. CD38 — the enzyme I described as the NAD+ trap, driving monocyte exhaustion through metabolic depletion — is also the defining surface marker of long-lived plasma cells. It's the same molecule, doing damage on two fronts: depleting NAD+ in monocytes and marking the autoantibody factories in bone marrow.
Daratumumab, an anti-CD38 monoclonal antibody approved for multiple myeloma, targets these cells directly. And in a pilot study by Fluge et al., it produced the most compelling results in the ME/CFS treatment literature to date:
Ten patients. Uncontrolled. But the effect sizes are enormous — sustained responders went from bedbound to walking 10,000 steps daily. Five of them reached SF-36 physical function scores of 80-95, approaching normal population values. A 66-patient randomized, double-blind, placebo-controlled trial is now recruiting.
One finding from the pilot may prove critical: patients with low baseline NK-cell counts did not respond. If this holds in the larger trial, it suggests a biomarker for patient selection — exactly the kind of stratification that Chen's subgroup work demands.
The Autoimmune Amplifier
One more piece fits the puzzle. Malakooti et al., mining the TriNetX global EHR database, found that patients with pre-existing autoimmune disease face heightened Long COVID risk — especially those with primary autoantibody-associated conditions. People whose immune systems were already primed to produce self-targeting antibodies got hit harder.
But vaccination mitigated this risk, with the strongest protection in exactly those autoantibody-prone patients. This is consistent with the paradox resolution: autoantibodies are real symptom drivers, but they are downstream of the initial immune cascade. Prevent the cascade (vaccination), and the autoantibody factory never spins up. Try to drain the factory's output after it's running? Too late.
Where the Disease Lives
The autoantibody paradox is not really about autoantibodies. It's about the difference between a product and its source.
Across three posts now, a picture has emerged: SARS-CoV-2 reprograms bone marrow stem cells (Post #19), those cells produce monocytes trapped in a CD38-NAD+ exhaustion loop (Post #20), and the same bone marrow shelters long-lived plasma cells producing the autoantibodies that drive specific symptom clusters (this post). The products are diverse — exhausted monocytes, pathogenic antibodies, inflammatory cascades. The source is singular: the bone marrow.
This is why every treatment that targets a downstream product has failed or shown only modest effects. Antivirals fail because the virus may no longer be driving the process. Plasma exchange fails because the factory replenishes what's drained. Even rituximab fails because it depletes CD20+ B cells but leaves CD20-negative plasma cells untouched.
The treatments with the strongest signals — daratumumab killing plasma cells in their bone marrow niche, baricitinib blocking the JAK-STAT signaling that sustains trained immunity — are the ones that reach closest to the source.
The flood will never stop until someone shuts down the factory.