Automated Prediction of Fibroblast Phenotypes Using Mathematical Descriptors of Cellular Features

Alex Khang,Abigail Barmore,Georgios Tseropoulos,Kaustav Bera,Dilara Batan,Kristi S Anseth
DOI: https://doi.org/10.1101/2024.05.15.594418
2024-05-18
Abstract:Fibrosis is caused by pathological activation of resident fibroblasts to myofibroblasts that leads to aberrant tissue stiffening and diminished function of affected organs with limited pharmacological interventions. Despite the prevalence of myofibroblasts in fibrotic tissue, existing methods to grade fibroblast phenotypes are typically subjective and qualitative, yet important for screening of new therapeutics. Here, we develop mathematical descriptors of cell morphology and intracellular structures to identify quantitative and interpretable cell features that capture the fibroblast-to-myofibroblast phenotypic transition in immunostained images. We train and validate models on features extracted from over 2,500 primary heart valve interstitial cells (VICs) and test their predictive performance on VICs treated with the small molecule drug 5-azacytidine, which inhibited myofibroblast activation. Collectively, this work introduces an analytical framework that unveils key features associated with distinct fibroblast phenotypes via quantitative image analysis and is broadly applicable for high-throughput screening assays of candidate treatments for fibrotic diseases.
Bioengineering
What problem does this paper attempt to address?