A. Cartagena-Rivera
National Institutes of Health (NIH),
United States
Keywords: mechanobiology, cell mechanics, atomic force microscopy, cancer
Summary:
The nanoscale surface mechanics of cells have been assessed numerous times for biomarkers of disease. However, cells are highly heterogeneous across their individual topographies and subcellular components, especially when compared to one another. High resolution Atomic Force Microscopy (AFM) methods can visualize these heterogeneities in a multi-dimensional (spatiotemporal) context, however processing such large datasets using a time-dependent viscoelastic approach represents a significant data science problem. Here, we introduce a method that leverages recent advancements in viscoelastic analysis via discrete modified Fourier transform (Z-transform). Our approach allows for the viscoelastic inversion of high-resolution spatiotemporal data at rates which are orders of magnitude faster (more than 1000 times) than optimizing a traditional rheological model for each pixel. In addition, the method utilized model-free viscoelastic quantities, such as the material retardance and relaxance. Next, we used machine learning to conduct unbiased analyses and classify the measured viscoelastic properties. Our nanoscale multi-timescale viscoelastic measurements revealed that 2D adherent metastatic melanoma cells exhibit reduction in elastic storage modulus and viscous loss modulus compared to melanocytes and fibroblasts. Moreover, we observed a progressive reduction in both storage and loss moduli with metastatic progression - a hallmark of cancer metastasis. Last, unbiased k-medoids clustering analysis revealed a previously unrecognized increase in nanoscale viscoelastic properties’ heterogeneity (quantified as an increase in observed clusters), possibly indicating increased structural heterogeneity of highly metastatic melanoma cells’ surface and cytoskeleton and increased cell migration. Altogether, the described method enables an efficient viscoelastic inversion of many experiments, creating a large dataset that can be easily passed to machine learning methods to dramatically improve the quality and quantity of biomarkers available to scientists, engineers, and biophysicists.