Xenium: the ligand-receptor pair that was segmentation leakage

The problem

The lab saw CD4 T cells that expressed both a chemokine normally made by dendritic cells and its receptor, and planned follow-up experiments on autocrine signaling. Expression was patchy and strongest where T cells touched dendritic cells.

What we found

• Default nuclear expansion of 5 µm in densely packed T-cell zones meant neighboring cells' transcripts were assigned to T cells.
• Mutually exclusive marker pairs (CD3E with ITGAX or LYZ) co-occurred in 14% of T cells, a direct measure of segmentation error.
• The 'autocrine' ligand transcripts sat at the cell periphery facing dendritic cells, not in the T-cell cytoplasm.

How we rescued it

1. Re-segmented with Proseg and compared it with Baysor, using membrane stain where available.
2. Quantified mixing before and after with exclusive-marker co-expression rates.
3. Re-annotated cells and re-ran neighborhood and ligand-receptor analysis restricted to true neighbors.

The result

Exclusive-marker co-expression dropped from 14% to under 3%. The autocrine signal disappeared, and a clear paracrine dendritic-cell-to-T-cell axis emerged at the T-B border. The lab redirected its validation experiments before spending months on the wrong hypothesis.