Machine learning identifies phenotypic profile alterations of human dopaminergic neurons exposed to bisphenols and perfluoroalkyls

Abstract

Abstract Parkinson’s disease (PD) is the second most common neurodegenerative disease and is characterized by the loss of midbrain dopaminergic neurons. Endocrine disruptors (EDs) are active substances that interfere with hormonal signaling. Among EDs, bisphenols (BPs) and perfluoroalkyls (PFs) are chemicals leached from plastics and other household products, and humans are unavoidably exposed to these xenobiotics. Data from animal studies suggest that ED exposure may play a role in PD, but data about the effect of BPs and PFs on human models of the nervous system are lacking. Previous studies demonstrated that machine learning (ML) applied to microscopy data can classify different cell phenotypes based on image features. In this study, the effect of BPs and PFs at different concentrations within the real-life exposure range (0.01, 0.1, 1, and 2 μM) on the phenotypic profile of human stem cell-derived midbrain dopaminergic neurons (mDANs) was analyzed. Cells exposed for 72 hours to the xenobiotics were stained with neuronal markers and evaluated using high content microscopy yielding 126 different phenotypic features. Two different ML models (XGBoost and LightGBM) were trained to classify ED-treated versus control mDANs. ED-treated mDANs were identified with high accuracy (0.92). Assessment of the phenotypic feature contribution to the classification showed that EDs induced a significant increase of alpha-synuclein (αSyn) and tyrosine hydroxylase (TH) staining intensity within the neurons. Moreover, microtubule-associated protein 2 (MAP2) neurite length and branching were significantly diminished in treated neurons. Our study shows that human mDANs are adversely impacted by exposure to EDs, causing their phenotype to shift and exhibit more characteristics of PD. Importantly, ML-supported high-content imaging can identify concrete but subtle subcellular phenotypic changes that can be easily overlooked by visual inspection alone and that define EDs effects in mDANs, thus enabling further pathological characterization in the future.