Large extracellular vesicles subsets and contents discrimination: the potential of morpho mechanical approaches at single vesicle level

Abstract Extracellular vesicles (EVs) are heterogenous lipid bound membranous structures released by different cells, showing a great potential to be used as biomarkers. They have also been explored for their role in the context of environmental toxicity. When endothelial cells are exposed to pollutants like Polycyclic Aromatic Hydrocarbons (PAH) - the most common being benzo[a]pyrene (B[a]P) – EVs released from those cells undergo surface and cargo modifications. Subpopulations of large EVs (lEVs) have shown to contain either damaged or intact mitochondria which is inexorably linked to oxidative stress conditions. In this paper, we studied B[a]P induced modifications in lEVs derived from endothelial cells, through morpho mechanical characterization with atomic force microscopy (AFM). Colocalizing AFM with fluorescence microscopy allowed us to differentiate between EVs containing mitochondria and those that did not. EVs containing mitochondria had a larger size (maximum diameter) when coming from treated cells (1.8 ± 0.89 µm) as compared to control cells (1.63 ± 0.76 µm). Moreover, their Young’s moduli were higher in the treated condition (3.09 ± 2.54 MPa in average) as compared to the control condition (1.25 ± 0.92 MPa in average). We also observed a heterogeneity within single vesicles, with most Young’s modulus values ranging from 0.1 up to 30 MPa for the treated condition and from 0.1 to 5 MPa for the control condition. Finally, applying linear discriminant analysis (LDA) and Random Forest (RF) algorithms on maximum diameter, height, and distribution of Young’s modulus values, we demonstrated the possibility to discriminate between EV subpopulations. Indeed, we successfully managed to a) distinguish EVs containing mitochondria from the “empty” ones, with an accuracy of 84% and b) discriminate whether these mitochondria-containing EVs originated from control or treated conditions, with an accuracy of 76%. These findings highlight the power of combining morpho-mechanical analysis and machine learning for identifying and discriminating EV subpopulations, no longer requiring any EVs fluorescence labelling. Competing Interest Statement The authors have declared no competing interest. Footnotes Additionnal supplementary figures, and one additionnal author