Kinetic Analysis of Label-Free Microscale Collagen Gel Contraction Using Machine Learning-Aided Image Analysis

Abstract

Pulmonary fibrosis is a deadly lung disease, wherein normal lung tissue is progressively replaced with fibrotic scar tissue. An aspect of this process can be recreated in vitro by embedding fibroblasts into a collagen matrix and providing a fibrotic stimulus. This work expands upon a previously described method to print microscale cell-laden collagen gels and combines it with live cell imaging and automated image analysis to enable high-throughput analysis of the kinetics of cell-mediated contraction of this collagen matrix. The image analysis method utilizes a plugin for FIJI, built around Waikato Environment for Knowledge Analysis (WEKA) Segmentation. After cross-validation of this automated image analysis with manual shape tracing, the assay was applied to primary human lung fibroblasts including cells isolated from idiopathic pulmonary fibrosis patients. In the absence of any exogenous stimuli, the analysis showed significantly faster and more extensive contraction of the diseased cells compared to the healthy ones. Upon stimulation with transforming growth factor beta 1 (TGF-β1), fibroblasts from the healthy donor showed significantly more contraction throughout the observation period while differences in the response of diseased cells was subtle and could only be detected during a smaller window of time. Finally, dose-response curves for the inhibition of collagen gel contraction were determined for 3 small molecules including the only 2 FDA-approved drugs for idiopathic pulmonary fibrosis.

Keywords: aqueous two-phase systems; collagen contraction; fibroblasts; machine learning; phenotypic assay; pulmonary fibrosis.