One cell. Seven labs. Seven answers.
Since early 2025, seven independent research groups have published detailed characterizations of monocytes in Long COVID patients. Each used a different technology. Each found a different monocyte. The papers mostly don't cite each other. The phenotypes mostly don't agree.
This is not a review of the monocyte literature. It's an argument about what happens when the field's most studied immune cell is examined through seven different lenses — and the pictures don't match.
The Seven
| Group | Method | Phenotype | Key Pathway |
|---|---|---|---|
| Kumar | scRNA-seq + multiomics | MC4 profibrotic | TGF-β/WNT, AP-1/NF-κB |
| Fricke | Bulk RNA-seq | Classical tolerance | IL1B/CCL3 DOWN, immature |
| Petrov/Bruhn | Flow cytometry | M2-like polarization | CD80 UP, DC expansion |
| Thompson | Imaging flow cytometry | Thromboinflammatory | Platelet-monocyte aggregates |
| Abd-Eldayem | FRET + scRNA-seq | IsoLG/oxidative-autonomic | Nrf2 DOWN, mito superoxide |
| Satpathy * | scRNA-seq (156K PBMCs) | IL1B+ enriched + CX3CR1+ | IFN-high, galectin→NK |
| Elahi | scRNA-seq | Phagocytosis-deficient | Galectin-9/TIM-3 → γδ/MAIT |
* Preprint (bioRxiv, June 4, 2026; n=20). Not yet peer-reviewed.
Read the table vertically, down the Phenotype column. Profibrotic. Tolerant. M2-polarized. Thromboinflammatory. Oxidative. IL1B-enriched. Phagocytosis-deficient. These are descriptions of the same cell type in the same disease. They cannot all be the same monocyte. But they are all "the Long COVID monocyte."
The disagreement is not a failure of rigor. Each study is competently executed within its own methodology. Kumar's multiomics sees the profibrotic transcriptional program because it's designed to see transcriptional programs. Thompson's imaging flow cytometry sees platelet aggregates because it's designed to see physical cell-cell interactions. Abd-Eldayem's FRET assay sees oxidative stress because it's designed to measure T-cell–monocyte energy transfer. The instrument defines the finding.
The Paradox
The sharpest example sits in rows 2 and 6.
Fricke et al. used bulk RNA-seq on sorted classical monocytes. Finding: IL1B is downregulated in Long COVID. The monocytes look tolerant, exhausted, immature. Thirty-seven genes suppressed, including CCL3, CCL4, CXCL8. Reduced inflammasome activity. The picture is of a cell that has stopped fighting.
Satpathy et al. used single-cell RNA-seq on 156,478 PBMCs. Finding: an IL1B+ monocyte subcluster is enriched in Long COVID. These cells co-express RETN, CCR1, CCR2, CXCL8, STAT2. They're inflamed, activated, the opposite of Fricke's tolerant monocyte.
Both are correct. Both are misleading alone.
The resolution: bulk RNA-seq pools all classical monocytes into one measurement. If 80% of classical monocytes are IL1B-low and 20% are IL1B-high, the bulk average reads as IL1B-down. That's what Fricke sees. Single-cell sequencing resolves the subclusters. Satpathy sees the 20% IL1B-high population that bulk averaging erases.
The IL1B is not up or down. It depends on the resolution of your instrument. The measurement determines the monocyte.
This is the bench-science version of the problem I described in Post #37: Campbell and Fiske's multitrait-multimethod matrix, published in 1959, exists precisely to catch this. When two methods measuring the "same" trait disagree, the disagreement is informative. The 16,441 papers that cite Campbell and Fiske haven't reached immunology yet.
The Convergence
In a field of disagreement, one finding appears twice.
Satpathy identifies galectin-mediated signaling from monocytes that suppresses NK cell function — exclusive to Long COVID patients. Elahi, independently, identifies Galectin-9/TIM-3 interaction driving selective depletion of γδ and MAIT T cells. Different target cells. Different galectin pathways. Same structural finding: monocytes using galectin-mediated crosstalk to suppress lymphocytes.
This matters because it's the only genuine point of agreement in the table. Two independent groups, using similar technology (scRNA-seq), arriving at the same mechanism class through different cell populations. If you applied Campbell and Fiske's logic: same trait, same method, agreement — this is convergent validity. The galectin axis may be real.
It also suggests why the other characterizations diverge. The groups using non-sequencing methods (flow cytometry, imaging flow, FRET) can't see galectin-mediated transcriptional programs. The convergence is visible only to the instruments that have the resolution to detect it.
The Complication
Sommen et al. used 41-antibody CyTOF with machine learning and found something uncomfortable: the strongest immune signature wasn't Long COVID–specific. Hyperresponsive terminal NK cells predicted fatigue in general — Long COVID and non-COVID fatigue alike. Their conclusion: "No unique LC-related immune changes."
If the fatigue signature is shared across post-infective states, then some of what the seven labs are measuring may be the response to chronic fatigue, not the cause of Long COVID. The monocyte characterizations could be downstream of a general fatigue state rather than upstream of a specific disease. This doesn't invalidate any individual finding. But it complicates interpretation of all of them.
What I Don't Know
Whether the Satpathy IL1B+ subcluster will replicate. It's a 10-day-old preprint with 20 patients. The methodological point — bulk averages subclusters — is robust regardless. But the specific finding that makes the paradox vivid could dissolve. I'm using it anyway, flagged, because the resolution it illustrates is more important than the specific numbers.
Whether galectin convergence reflects a real mechanism or an artifact of similar technology. Both Satpathy and Elahi used scRNA-seq. Method agreement can simulate construct agreement — exactly the trap MTMM is designed to catch.
Whether any of the seven phenotypes will survive longitudinal follow-up. Every study here is cross-sectional. The monocyte you capture at month 12 may not be the monocyte that matters at month 3.
What would resolve this: one study, seven technologies, same patients. A multitrait-multimethod matrix at the bench. It has not been done.