Kintsugi maps nucleus-poor RNA compartments in subcellular spatial transcriptomics

Spatial transcriptomics enables the resolution of RNA within tissue structures that extend beyond nuclei. However, existing aggregation methods either impose fixed regions or use cells as anchors, making it challenging to directly measure process-rich and extracellular compartments. To address this, researchers have developed Kintsugi, a deterministic tessellation method that partitions sparse count matrices based on captured UMI density and assigns every bin within the tissue.

By separating Pearson-residual gene composition from captured transcript density, Kintsugi preserves complete tissue representation without the need for histology or nuclear segmentation.

In mouse brain Visium HD data, Kintsugi successfully identified nucleus-poor neuropil compartments enriched with glial and dendritic transcripts. Additionally, Xenium molecule coordinates supported the localization of dendritic mRNA away from nuclei. The same tissue representation also identified nucleus-poor fibrotic scar regions in idiopathic pulmonary fibrosis and matrix structure in fetal cartilage. Furthermore, CODEX proteomics revealed that the captured density of RNA was not reducible to nuclear packing alone.

These findings demonstrate that nucleus-poor tissue spaces contain structured and interpretable RNA biology. Through complete tissue tessellation using Kintsugi, researchers can now access and study the RNA biology present in these previously challenging compartments.