Quantitative Image Analysis of Fibrillar Collagens Reveals Novel Diagnostic and Prognostic Biomarkers and Histotype-Dependent Aberrant Mechanobiology in Lung Cancer

Abstract

Fibrillar collagens are the most abundant extracellular matrix components in non–small cell lung cancer (NSCLC). However, the potential of collagen fiber descriptors as a source of clinically relevant biomarkers in NSCLC is largely unknown. Similarly, our understanding of the aberrant collagen organization and associated tumor-promoting effects is very scarce. To address these limitations, we identified a digital pathology approach that can be easily implemented in pathology units based on CT-FIRE software (version 2; https://loci.wisc.edu/software/ctfire) analysis of Picrosirius red (PSR) stains of fibrillar collagens imaged with polarized light (PL). CT-FIRE settings were pre-optimized to assess a panel of collagen fiber descriptors in PSR-PL images of tissue microarrays from surgical NSCLC patients (106 adenocarcinomas [ADC] and 89 squamous cell carcinomas [SCC]). Using this approach, we identified straightness as the single high-accuracy diagnostic collagen fiber descriptor (average area under the curve = 0.92) and fiber density as the single descriptor consistently associated with a poor prognosis in both ADC and SCC independently of the gold standard based on the TNM staging (hazard ratio, 2.69; 95% CI, 1.55-4.66; P