Cytokine release syndrome (CRS) is a dose-limiting toxicity of T-cell engagers and agonistic antibodies. Preclinical assessment still relies almost entirely on measuring inflammatory cytokines in culture supernatants. Those readouts show what cytokines are released, but not which cells release them, or through which transcriptional programs — critical data to understand and mitigate the toxicity.
Resolving it at single-cell level across many conditions, timepoints and donors is where conventional droplet-based scRNA-seq becomes too expensive. With the STAMP approach, Xenium can profile dissociated cells with a targeted in situ panel. This decouples cost from cell number and makes large factorial designs practical.
Study design
- Donors: peripheral blood mononuclear cells (PBMCs) from 3 healthy donors
- CRS-inducing agents: blinatumomab (CD19×CD3 bispecific T-cell engager), tarlatamab (DLL3×CD3 bispecific T-cell engager) and TGN1412 (CD28 superagonist)
- Non-CRS-inducing control: pembrolizumab (anti-PD-1)
- Timepoints: 6, 24 and 48 h, alongside an untreated T0 control
- Functional readout: supernatants assayed by cytometric bead array (CBA) for secreted cytokines
- Single-cell readout: cells from all 39 conditions deposited on a single Xenium slide and profiled with a 378-plex panel, yielding about 1 million high-quality cells in one run
- Analysis: 24 cell populations annotated across nine lineages, including activation, memory and exhaustion states. Pseudobulk differential expression was run per sample, lineage and timepoint (edgeR LRT). Hallmark pathway activity was scored by limma, with contrasts against both pembrolizumab and untreated control.
What STAMP revealed
The cellular source of released cytokines
CBA confirmed the model. IL-6, TNF-α, IL-1β and IL-10 rose sharply from 24 h onwards, and only under the three CRS-inducing agents. STAMP then traced these signals to their cellular source. IL10 expression mapped to a discrete cluster of activated CD4+ T cells. This held across all three CRS-inducing agents, and the cluster was absent under pembrolizumab.
A shared IL6–JAK–STAT3 circuit with a hidden discordance
At the mechanistic level, the three agents converged on a shared transcriptional program organized around two cytokine circuits. The first was the IL6–JAK–STAT3 axis. LIF and CSF1 were induced across most lineages, while IL6 transcript itself was downregulated. This points to gp130 signaling driven by alternative ligands, a discordance that protein measurement alone cannot detect.
An ordered IL2–STAT5 cascade
The second circuit, the IL2–STAT5 axis, followed an ordered cascade. IL2 was induced transiently in CD4+ T cells at 6 h. Its receptor chains IL2RA (CD25) and IL2RB were then induced in a sustained way at 24–48 h. TNFRSF9 (4-1BB) marked engagement from 6 h across all nine lineages. Proliferation was a late, pan-lineage event at 48 h, extending to lineages the engagers do not directly target. Pembrolizumab stayed near baseline throughout, for both circuits.
Why it matters for drug developers
Pairing a functional cytokine readout with targeted single-cell transcriptomics turns a quantitative measurement into a mechanistic one. It shows what is released, by which cell type, and through which program. Xenium STAMP makes this practical: over a million cells in a single run, at a cost and turnaround that standard single-cell workflows cannot match for designs of this size.
The poster illustrates the approach with a CRS model, but it applies to any cell-based assay currently limited to functional endpoints, such as T-cell killing, MLR or macrophage assays.
Authors
Jean-Philippe Guégan, Olivier Mauduit, Christophe Rey, Lucie Lefèvre, Oren Lara, Corentin Pignon, Ophélie Odin, Assia Chaibi, Imane Nafia, Alban Bessède
Explicyte Immuno-Oncology, Bordeaux, France
