Abstract Background ATP-binding-cassette-transporter-A3 (ABCA3) transports phospholipids across the lysosomally-derived, lamellar body membranes in ATII cells to assemble surfactant. Biallelic pathogenic ABCA3 variants are the most common genetic cause of surfactant dysfunction. Current therapies are limited and non-specific. More than 300 disease-associated ABCA3 variants have been identified, and most are rare/private. Fewer than 10% have been functionally characterized with primary assays (colocalization, immunoblotting, ABCA3+ vesicle diameter), which are time-consuming and not scalable. Currently recognized pathogenic mechanisms include disruption of intracellular trafficking or impairment of ATP-mediated phospholipid transport. The encoded mechanism of an ABCA3 variant is difficult to predict based on the location in the gene/protein, and accuracy of in silico pathogenicity prediction algorithms remains limited. Objective To use high content analysis (HCA)/deep cellular phenotyping for rapid mechanistic classification of disease-associated ABCA3 variants. Methods We used a human pulmonary epithelial cell line (A549) with genetically abrogated endogenous ABCA3 expression (A549/ABCA3-/-) which contains a LoxFas/LoxP landing pad cassette that permits stable expression of individual ABCA3-mCherry variant constructs from a single intergenic location. We used 9 cell lines that stably express either wild-type (WT) ABCA3-mCherry or individual disease-associated ABCA3-mCherry variants, including previously characterized mistrafficking (severe, partial) and phospholipid transport variants. To eliminate batch effects and limit technical biases, we plated cells in a semi-random pattern across multiple 96-well plates, stained with Hoechst (nucleus) and LysoTracker (lysosome-related organelles), and performed HCA imaging (Molecular Devices HTai high-content confocal microscope, 20x magnification). Tracing and feature extraction were performed using Cytiva’s InCarta software and used to train artificial neural networks. We used classifiers containing multiple fluorescence-based cellular features, uniform-manifold-approximation-projection (UMAP) dimensional reduction, and Kruskal-Wallis test to compare cells that express WT-ABCA3-mCherry or ABCA3 variants. Results Nine fluorescence-based cellular features describing ABCA3 intensity and spatial distribution were sufficient to distinguish between severe mistrafficking (W179C,M760R,D1058Y,N1076K), partial mistrafficking (R280C,K1388N,G1421R), or phospholipid transport (E292V) variants (KW p = 4.38x10-22, Fig. 1). Using these features, individual round-robin validation models achieved accuracies of 60-86%, indicating robust variant discrimination. Discussion Functional characterization results with HCA were similar to established functional primary assays and were more rapid and efficient (8 variants screened in 3 days vs. months for primary assays). Cellular features detected with HCA can be used for rapid mechanistic classification of ABCA3 variants. This approach may be adapted to high-throughput screening of compounds for correction of ABCA3 variant-encoded surfactant dysfunction and inform development of therapeutic modulator or potentiator strategies similar to those developed for cystic fibrosis. This abstract is funded by: NHLBI, NCATS, Children’s Discovery Institute
Wambach et al. (Fri,) studied this question.