STITCH links cellular morphology and gene expression in spatial transcriptomics

Area of Science:

• Computational Biology
• Spatial Transcriptomics
• Bioinformatics

Background:

• In situ spatial sequencing (ISS) links cellular morphology and gene expression in vivo.
• Existing methods, particularly deep learning, confound cell shape and size.
• A principled, interpretable mathematical representation of cell morphology is needed for ISS.

Purpose of the Study:

• To present an interpretable, size-independent representation of cellular boundary contours.
• To develop STITCH (Shape-TranscriptomIc Correlation and Harmonization) for morphology-transcriptome analysis in ISS datasets.
• To demonstrate STITCH's ability to recover known and discover novel morphology-gene expression relationships.

Main Methods:

• Tangent Principal Component Analysis (TPCA) on a Kendall shape manifold for size-independent contour features.
• Development of the STITCH algorithm integrating TPCA with gene expression data.
• Application of STITCH to Xenium and CosMx spatial transcriptomics datasets.

Main Results:

• TPCA successfully recovers shape-perturbing genes, outperforming previous methods.
• STITCH outperforms deep learning in identifying keratinocyte organization and nuclear eccentricity gradients in Xenium data.
• STITCH links fibroblast morphology to malignant cell proximity and myofibroblast programs in CosMx melanoma data.
• STITCH independently confirms links between mesenchymal cell states and cell area in melanoma.

Conclusions:

• STITCH provides an interpretable mathematical framework for morphology-transcriptome relationships in spatial biology.
• The method accurately captures size-independent cell shape features for biological interpretation.
• STITCH advances the analysis of complex spatial transcriptomics data across various cell types and platforms.