A reference-calibrated spatial-statistical framework for quantifying stroma-tumor cell spatial associations from histology and its geometric limits

Quantifying interactions within biological tissues necessitates identifying whether an interaction exists, the distance at which it occurs, and its intensity. While existing spatial-statistical methods can illustrate spatial organization, they typically necessitate specifying an interaction range or model beforehand, making them less suitable for scenarios where these parameters remain unknown.

This study introduces a pragmatic framework that ascertains the existence, reach, and association intensity of spatial bias towards a tissue structure directly from segmented histology. The framework amalgamates four distinct statistics – the resource selection function, nearest-structure distance distributions, the cross-J-function, and the radial distribution function – with specimen-specific reference states that maintain tissue stromal architecture while traversing a spectrum of known spatial biases.

The findings from controlled in silico tissues reveal that tissue stromal architecture imposes geometric constraints on the range of recoverable interactions, and the event fraction independently determines statistical sensitivity. These spatial associations establish essential boundaries for inference and underscore the importance of interpreting spatial statistics relative to tissue-specific reference states rather than fixed scales.

The framework was applied to proliferating tumor cells in relation to CAF-rich stroma during tyrosine kinase inhibitor therapy in an ALK+ NSCLC model, unveiling a significant stroma-associated proliferation bias extending roughly 30-60 micrometers. This framework is versatile and can be applied to any histological context where events and reference structures can be defined, offering a practical method to quantify localized interactions directly from tissue stromal architecture.