Use of machine learning for quantification of retinal pigment epithelium tight junctions improves assay sensitivity

Abstract

The retinal pigment epithelium (RPE) is critical for maintaining outer retinal barrier homeostasis. In age-related macular degeneration (AMD), the RPE can undergo a dedifferentiation process that includes tight junction (TJ) loss and displacement of zonula occludens-1 (ZO-1), which may impair structural and functional integrity of the RPE barrier and contribute to disease pathogenesis. Our objective was to develop an automated and sensitive quantification method for TJ aberrations in an RPE immunofluorescence imaging assay, following treatment with TNF or TGF{beta}2. However, quantifying ZO-1 morphological changes in the RPE using standard image analysis methods did not provide a satisfactory assay window. To address this challenge, we developed an imaging assay to quantify ZO-1 changes using a machine learning approach, enabling enhanced phenotypic characterization of the ZO-1 changes in RPE cells and improved assay sensitivity. We were also able to capture and quantify the reversal of these changes using etanercept, an TNF inhibitor, with this imaging assay. Our findings indicated that this machine learning ZO-1 quantification assay could serve as a potential phenotypic readout for RPE dedifferentiation and enabling large-scale mechanistic studies.