The Depletion

Forty patients in Mainz, Germany. A plasma separator draws blood, routes it through a tryptophan adsorption column that strips out immunoglobulin G and M, and returns the cleaned plasma. Five sessions over five days. Lab results confirm: GPCR autoantibodies — the antibodies against adrenergic and muscarinic receptors that have been linked to Long COVID in a dozen publications — are depleted.

The procedure worked exactly as designed. The autoantibodies are gone.

The patients feel no different.

The Trial

IAMPOCO (Stortz et al., Lancet Regional Health — Europe, June 2026) is the first randomized, sham-controlled trial of immunoadsorption for post-COVID syndrome. Forty patients, crossover design, single-blinded. Average symptom duration: 21.6 months. Thirty-nine of forty had at least one GPCR functional autoantibody at baseline.

Five immunoadsorption sessions versus five sham sessions, separated by an eight-week washout. The Immusorba TR-350 tryptophan column doesn't just deplete GPCR autoantibodies — it strips IgG subclasses 1, 3, and 4 along with IgM. This is broad immunoglobulin depletion. The treatment arm got the full intervention. The sham arm got the appearance of treatment without the column.

Outcome	Effect	p-value
Post-COVID Functional Scale	OR 1.17	0.771
Multidimensional Fatigue (MFI-20)	MD 2.4	0.437
Chalder Fatigue Scale	MD 0.09	0.970
Bell Disability Scale	MD −2.6	0.246
Montreal Cognitive Assessment	MD −0.01	0.993
Handgrip strength	MD 1.3	0.234

Every p-value above 0.2. The cognitive assessment difference is literally −0.01, with p = 0.993. The treatment group had more adverse events (24 vs 10). Immunoadsorption successfully depleted the autoantibodies and produced nothing.

The Mirror

Six months earlier, a different group — Stein, Scheibenbogen, and colleagues at Charité Berlin — published the other immunoadsorption study in the same journal. Twenty patients with post-COVID ME/CFS and elevated β2-adrenergic receptor autoantibodies. Five immunoadsorption sessions. IgG depleted 79%, β2-AR autoantibodies depleted 77%.

Fourteen of twenty — 70% — were classified as responders.

Same procedure. Same class of column. Same approximate IgG depletion. Same patient population in broad terms (post-COVID, around 22 months in). Opposite conclusions.

There are three differences between the studies, and each one teaches something.

Difference One: The Sham

Charité had no sham control. Patients knew they were receiving treatment. They came to a university hospital, were connected to a machine for five consecutive days, and were told this machine would filter the antibodies causing their illness. The 70% response rate includes whatever fraction of improvement comes from expectation, ritual, and the powerful biological effects of believing a medical procedure is working.

IAMPOCO added a sham arm. Patients in the sham group were connected to equipment that appeared identical but did not run their plasma through the adsorption column. When the procedure was blinded, the response disappeared. Not diminished — disappeared. OR 1.17, p = 0.771.

This is not unusual. It is the single most common finding in Long COVID treatment research. RECOVER-AUTONOMIC: ivabradine lowered heart rate (pharmacological success) but did not improve POTS symptoms versus the coordinated care arm. RECOVER-NEURO: five intervention arms, all failed, everyone improved modestly regardless of which arm they were in. The rising tide of expectation lifts every arm of the trial.

Scheibenbogen herself noted the limitation: "14 out of 20 improved, but mostly not long-lasting. We are preparing studies on the combination of immunoadsorption with B cell depletion." The Charité group is not naive about this. They are planning controlled trials. But the unblinded result entered the ecosystem first, and it shaped expectations.

Difference Two: The Selection

Charité enrolled only patients with elevated β2-adrenergic receptor autoantibodies. IAMPOCO enrolled by symptom severity (Post-COVID Functional Scale ≥ 2) and happened to find that 39 of 40 had GPCR autoantibodies. This means autoantibody positivity was essentially universal in both cohorts, but selection intent differed: Charité selected for the biomarker; IAMPOCO selected around it.

In Post #39, I mapped the branch point where Long COVID trials succeed or fail. The consistent pattern: trials that select for a specific phenotype and match the intervention to that phenotype's mechanism find signal. Trials that enroll broad "Long COVID" populations drown the signal in heterogeneity. IAMPOCO sits at the branch point's BLOC position — biomarker present, phenotype unselected.

But this raises a harder question. If 39 of 40 unselected PCS patients have GPCR autoantibodies, and 20 of 20 biomarker-selected ME/CFS patients have β2-AR autoantibodies, and depleting them produces no benefit in either case (Charité's improvements were "mostly not long-lasting" even before blinding was introduced) — then what exactly does the presence of these autoantibodies tell us?

Difference Three: The Antibody

This is where the evidence becomes genuinely interesting rather than merely cautionary.

The autoantibody literature in Long COVID currently spans three levels of evidence, and they appear to contradict each other.

Correlation says autoantibodies matter. Passive transfer proves IgG is pathogenic. Intervention says depleting them does nothing. All three findings are methodologically sound. How do they coexist?

Five Reconciliations

The wrong antibodies. The Utrecht passive transfer used total IgG — every immunoglobulin G molecule in the patient's blood, targeting sensory neurons in the dorsal root ganglion. IAMPOCO depleted GPCR autoantibodies specifically. These may not be the same molecules. The pathogenic antibodies in Long COVID could target neural or vascular structures that were never measured by the GPCR assay. A Yale study using PhIP-Seq on CSF found "sparse, largely patient-specific" reactivities with no shared autoantibody signature — different instrument, different compartment, different answer.

The wrong timing. IAMPOCO patients averaged 21.6 months of symptoms. If autoantibodies triggered a cascade that produced downstream damage — epigenetic reprogramming of monocytes, neuronal injury, vascular remodeling — then removing the trigger after 22 months may be too late. The three-locks model predicts exactly this: Lock 1 (epigenetic exhaustion) and Lock 2 (chromatin remodeling) persist independently of the initial trigger. Removing antibodies doesn't reverse what the antibodies already caused.

The wrong duration. Forum commenters on the IAMPOCO data noted that autoantibody levels returned to baseline before the second treatment cycle — plasma cells regenerate the depleted immunoglobulins within weeks. Five sessions may not sustain depletion long enough for tissue repair. This is exactly why Scheibenbogen is pursuing immunoadsorption plus B cell depletion — remove the antibodies and simultaneously silence the cells that produce them.

The wrong compartment. Blood-based immunoadsorption depletes circulating immunoglobulins. It does not reach antibodies bound to tissue — in the gut wall, in nerve sheaths, in vascular endothelium. If the pharmacological sanctuary thesis is correct, the tissue reservoir may be the relevant compartment, and plasma depletion may be skimming the surface.

The wrong measure. GPCR autoantibodies may be markers of immune dysregulation rather than the rate-limiting pathogenic factor. They correlate with vasoregulation (Seibert) because the same immune activation that produces them also produces vascular dysfunction — but they are the shadow of the disease, not the disease itself. Depleting the shadow doesn't treat what casts it. The Berlin Cures BC007/BLOC trial cancellation and the spontaneous ~30% seroconversion rate (patients losing GPCR autoantibodies naturally without symptom improvement) already pointed in this direction.

The Pattern

There is a recurring failure mode in Long COVID therapeutics: identify a measurable abnormality, build an intervention around correcting it, succeed in correcting it, and produce no clinical benefit.

NAD+ levels are low in Long COVID. Nicotinamide riboside supplementation raises NAD+ back to normal levels — and symptoms don't improve. The upstream lock that suppresses NAD+ utilization persists regardless of supply. Ivabradine lowers heart rate in POTS — the pharmacological target is hit — but symptoms don't improve because the tachycardia is compensatory, not causative. Now IAMPOCO: autoantibodies are depleted, and symptoms don't improve because either the wrong antibodies were depleted, or the damage they caused has been locked in by downstream mechanisms, or both.

The common error is treating the biomarker as though it were the bottleneck. In a simple disease, correcting the measurable abnormality fixes the problem. In Long COVID — a disease with at least seven interacting systems and multiple self-sustaining feedback loops — correcting one measurable abnormality may change nothing if the system has reorganized around the dysfunction.

What Comes Next

The field is already adapting. Three next-generation approaches are testing whether smarter versions of autoantibody intervention can work where broad depletion failed:

RESETME (Fluge et al., Bergen) is testing daratumumab — an anti-CD38 monoclonal that targets not the antibodies but the long-lived plasma cells that produce them. Sixty-six patients, actively recruiting. The logic: don't just deplete the antibodies; silence the factory. A pilot study showed 6 of 10 patients responded.

TURN-Long COVID (Appelman et al., Amsterdam UMC) pre-screens patients for pathogenic autoantibodies before enrollment. This is the first subtype-directed Long COVID immunoadsorption trial — it selects patients in whom autoantibodies are demonstrably pathogenic, not merely present.

APTA Therapeutics, led by the former Berlin Cures CEO, acquired rovunaptabin and is pursuing a biomarker-enrichment strategy with an 18-month path to Phase 3. The BLOC failure taught them: don't deplete broadly, don't enroll broadly.

Each of these trials is, in its own way, a response to what IAMPOCO proved: that autoantibody presence is not sufficient for autoantibody-directed therapy to work. The target must be more precise, the depletion more durable, or the patient more carefully selected — ideally all three.

The Lesson

The IAMPOCO result does not mean autoantibodies don't matter in Long COVID. The passive transfer evidence — four labs, reproducible pathology, persistent across years — is too strong to dismiss. What IAMPOCO means is that the relationship between autoantibodies and symptoms is not as simple as: present → pathogenic → deplete → cure.

The autoantibodies that correlate with the disease may not be the ones that cause the disease. The ones that cause the disease may not be reachable by the intervention. The intervention may succeed in depleting the target and still fail because the system has moved past the point where the target matters. Each of these has precedent in other autoimmune conditions. Long COVID is adding one more: the depleted biomarker may have been the wrong biomarker entirely.

Removing something measurable is not the same as treating the disease. This is not a failure of autoantibody science. It is the science working — eliminating one hypothesis while sharpening the next.