PTEC Morphology

Bright-field images documenting PTEC-TERT1 morphological changes in Figure 1 were taken at day 9 of AON treatment on living cells at 20× magnification.

For high-content imaging of cell morphology, PTEC-TERT1 cells were seeded on CellCarrier-96 collagen-I-coated microplates (PerkinElmer, catalog #6005920). Cell monolayers were fixed with 4% paraformaldehyde in PBS for 20 min at room temperature. Cells were permeabilized and blocked using 0.1% Saponin (Sigma-Aldrich, catalog #47036) and 4% gelatin from cold-water fish skin (Sigma-Aldrich, catalog #G7041) in PBS for 10 min at room temperature. ActinGreen 488 ReadyProbes Reagent (Thermo Fisher Scientific, catalog #{"type":"entrez-nucleotide","attrs":{"text":"R37110","term_id":"794566","term_text":"R37110"}}R37110) was applied according to the manufacturer’s instructions. DRAQ5 (Thermo Fisher Scientific catalog # 62251) was used for nuclear and cytoplasmic staining according to the vendor’s instructions. Images were acquired on the Operetta High-Content Imaging System (PerkinElmer). Fluorescence images were taken from nine fields per well (central area, 3.1 mm²) at 20× magnification.

To quantify AON-induced changes in cell morphology, high-content image information was extracted and analyzed using the Columbus Image Data Storage and Analysis software (Operetta Application Guide 2013, PerkinElmer). Computational workflow included, in the following order: (1) identification of primary objects (cell nuclei and cytoplasm) based on DRAQ5 staining; (2) exclusion of cropped cells at image borders; and (3) sequential calculation of morphological properties based on two software building blocks—STAR morphology, assessing the outer shape of objects and fluorescence intensity distribution inside the objects, and SER texture properties, representing the regularity of intensities inside the objects. Phenotypes of interest were quantified using the PhenoLOGIC machine learning module of the Columbus Image Analysis software, which classifies the cells based on the pre-calculated STAR and SER features. The algorithm was trained by interactive machine learning on two cell populations: vehicle-treated and AON-C (100 μM)-treated cells. For each population of vehicle- and drug-treated cells, at least 100 training objects were randomly selected over the whole image area. The algorithm identified a linear combination of properties that best separated the training samples. Data are expressed as percentage of cells retaining normal, vehicle-like morphology.