Spatial Topology Reveals Biologically Distinct Recurrent Motifs in Colorectal Cancer

Area of Science:

• Computational Biology
• Cancer Research
• Spatial Transcriptomics

Background:

• Current spatial transcriptomic analyses often rely on predefined cell types or compartments.
• Discovering recurrent multicellular tissue architectures across patients is challenging.

Purpose of the Study:

• To develop an unsupervised method for identifying recurrent spatial motifs in solid tumors.
• To investigate the role of identified motifs in colorectal cancer, particularly early-onset disease.

Main Methods:

• Developed STORM (Spatial Topology analysis of Recurrent Motifs), an unsupervised graph-attention variational autoencoder.
• Applied STORM to Xenium spatial transcriptomic data from early-onset and average-onset colorectal cancer patients.
• Utilized Vision Transformer for translating motif macro-classes to H&E images.

Main Results:

• Identified 10 recurrent spatial motifs, including the Desmoplastic Fibrotic Barrier (DFB).
• The DFB motif restricts CD8+ T-cell access to tumor cores, independent of T-cell abundance.
• Early-onset colorectal cancer showed amplified DFB architecture and an associated age-specific prognostic signature.
• An image-derived DFB-barrier composite predicted overall survival in advanced colorectal cancer.

Conclusions:

• STORM offers an annotation-free framework for discovering recurrent spatial motifs in tumors.
• A fibroblast barrier architecture (DFB) is identified as a key player in immune exclusion.
• This DFB architecture is amplified in early-onset disease and has prognostic value, translatable to pathology images.