Integrating histopathology and transcriptomics for spatial tumor microenvironment profiling in a melanoma case study.

Area of Science:

• Computational pathology
• Cancer research
• Bioinformatics

Background:

• Tumor microenvironment (TME) cellular structures influence tumor progression and treatment outcomes.
• Quantitative analysis of TME architecture from standard histopathology slides is challenging.
• Existing methods may not fully capture the spatial complexity of cellular interactions.

Purpose of the Study:

• To introduce SPoTLIghT, a novel computational framework for quantitative analysis of tumor architecture using H&E slides.
• To develop a method for generating spatial cellular maps and deriving interpretable features of TME organization.
• To demonstrate the utility of SPoTLIghT in identifying TME subtypes, understanding immune infiltration, and predicting patient prognosis.

Main Methods:

• Development of a weakly supervised machine learning model trained on melanoma patient data.
• Extraction of tile-level imaging features from H&E slides linked to RNA-sequencing data.
• Generation of spatial cellular maps and conversion into graphs to derive 96 interpretable features.
• Application of SPoTLIghT to distinguish TME subtypes and predict prognosis in an independent cohort.

Main Results:

• SPoTLIghT successfully generates spatial cellular maps from H&E images.
• Derived features capture TME cellular organization and distinguish microenvironment subtypes.
• The framework reveals nuanced immune infiltration patterns beyond molecular data.
• SPoTLIghT accurately predicts patient prognosis in an independent melanoma cohort.

Conclusions:

• SPoTLIghT provides a quantitative and interpretable characterization of tumor spatial contexture.
• The framework enhances computational histopathology by integrating imaging and molecular data.
• SPoTLIghT offers a valuable tool for understanding TME complexity and improving cancer patient outcomes.