Evaluating cytoskeleton integrity by fast-AFM and supervised machine learning

The cytoskeleton is an underlying protein network supporting the cell membrane and its integrity is crucial for the Red Blood Cell (RBC) deformability, which in turn mediates the function of these cells. In the particular context of infectious models, the extent of defomability of the RBCs can be linked to the likelihood of invasion, as in the case of the human malaria parasite. These deadly parasites secrete Extracellular Vesicles (EVs), while growing inside the RBCs. There is growing evidence that RBC deformability is impaired in malaria invasion. However, the mechanistic role of parasite-derived EVs on the RBC host membrane is not yet understood.

Here, we have applied atomic force microscopy (AFM) to study the mechanical changes occurring in RBCs treated with malaria-derived EVs, as well as morphological transformations in the cellular cytoskeleton. High-resolution images of dried cells with exposed cytoskeleton show distinct morphological differences associated with the breakdown and softening of the cell structure.

Further, a deep learning model (convolutional neural network) was applied in order to verify and quantify the differences between images of healthy and damaged cytoskeleton. Importantly, the model also provided an independent test of the efficacy of drug treatment for prevention of EV-induced damage. In order to acquire the large volume of images necessary for training the model and to obtain sufficient statistics, images were obtained using a newly designed fast-scanning AFM system.