In November 2024, Berlin Cures announced that BC 007 — a synthetic DNA aptamer designed to neutralize autoantibodies against G-protein coupled receptors — showed no superiority over placebo in its Phase 2 BLOC trial. The company went bankrupt. The drug appeared dead.
Five months later, the reCOVer trial at University Hospital Erlangen published in eClinicalMedicine. Same drug. Same target. Positive results: Bell Disability Scale effect size 3.64 (p=0.0004), FACIT fatigue 2.10 (p=0.038), FSS -2.66 (p=0.009). Four of eight SF-36 quality-of-life domains reached significance.
Same molecule. Opposite conclusions. The tree above explains why.
The Denominator, Not the Drug
Both trials required GPCR-fAAb positivity. Both used intravenous rovunaptabin. Both were placebo-controlled and double-blinded. The difference was the denominator.
BLOC enrolled 119 patients across 14 centres in five countries. Its inclusion criteria: FACIT fatigue score below 35 plus at least one additional persistent symptom. That’s a wide net within the GPCR-fAAb-positive population. Any fatigue, any one symptom, any centre.
reCOVer enrolled 30 patients at a single centre. Its criteria: Bell score ≤60 (more severe disability), at least three of eight defined symptoms, fatigue as the major symptom, and — critically — exclusion of patients with organ damage. The population was narrower at every level.
The tree narrows from top to bottom. At each branch point, some patients are included and some are excluded. The treatment effect appeared only at the most specific branch — not because the drug changed, but because the population did.
Depth Alone Isn’t Enough
A simple reading of BC007 would be: select harder, and trials work. But the tree has another branch that disproves this.
GeNeuro’s temelimab trial (n=203) selected patients by W-ENV biomarker — a different autoantibody target linked to HERV-W envelope protein reactivation. Biomarker-enriched. Still failed. The selection was precise, but the mechanism was wrong, or the biomarker didn’t correspond to the pathology driving symptoms.
Selection is necessary. It is not sufficient. You need the right branch and the right depth.
The Measurement Wrinkle
There’s a detail in the reCOVer data that complicates even the positive result. Thirty percent of patients — nine of thirty — became GPCR-fAAb seronegative spontaneously during the trial, without receiving rovunaptabin. Their autoantibodies dropped below detection threshold on their own.
They showed no fatigue improvement.
If the autoantibodies are causing the fatigue, why didn’t their disappearance fix it? Two possibilities. First, sub-detection-threshold autoantibodies may still drive pathology — the assay’s sensitivity floor defines the construct it measures, not the biology. Second, by the time autoantibodies drop, downstream damage (autonomic, vascular, neurological) may already be self-sustaining.
Either way, the bioassay doesn’t perfectly circumscribe the disease. The branch on the tree has finer subdivisions we can’t yet see.
The Industry Learned
The most remarkable thing about the BC007 story isn’t the divergent results. It’s what happened after.
Berlin Cures went bankrupt in November 2024. Within months, a new company — APTA Therapeutics, founded by Berlin Cures’ former CEO Oliver von Stein — acquired rovunaptabin through Ascenion, the tech transfer office for Charité and the Max Delbrück Center. APTA’s stated strategy: biomarker-enrichment from the start. An 18-month path to Phase 3-quality data, explicitly correcting the enrollment error that sank BLOC.
The field is watching the tree and learning where to stand on it.
A Lancet Infectious Diseases editorial framed the broader lesson: phenotypic heterogeneity in Long COVID risks eclipsing responder subgroups in trials that don’t account for it. Negative results don’t mean the drug failed. They may mean the denominator was wrong.
What the Tree Shows
I’ve been building toward this argument across twelve posts. The pattern repeats:
- RECOVER-AUTONOMIC: ivabradine lowered heart rate but missed the primary endpoint. The category “POTS” was too broad.
- RECOVER-NEURO: five cognitive interventions, all null, all improved equally. 60.9% of enrollees had no objective impairment.
- MTMM analysis: reliable instruments measuring incoherent constructs. The tools work. The categories don’t.
BC007 is the cleanest case because it’s a matched pair: same drug, same mechanism, different denominator, opposite result. The category didn’t kill the drug. The drug is fine. The category killed the trial.
The lesson isn’t for Long COVID alone. Any disease defined by a symptom cluster rather than a mechanism will face this. The label organizes funding, trials, advocacy, and clinical practice around a unity that doesn’t exist biologically. Every trial that enrolls “Long COVID patients” without specifying which Long COVID is standing at the trunk of the tree, hoping the branches don’t matter.
They do.