Cellular Mechanical Behavior Study and Cytoskeleton Characterization

Abstract

Mechanotransduction–the process living cells sense and respond to forces–is essential for the maintenance of normal cell, tissue, and organ functioning. To promote the knowledge of mechanotransduction, cell modeling plays a key role in unveiling the cellular force sensing and transduction mechanism. However, most studies either treat the cells as a homogeneous elastic or viscoelastic material, which is far from the real structure of cells, or need to run experiments manually under nanoscale, which is still laborious and challenging. In this dissertation, multiple models are developed to improve the investigation accuracy and efficiency. In Chapter 2, we propose a model using the finite element method (FEM) to simulate the force response of living cells during the indentation from an atomic force microscope (AFM). The model consists of a multi-layered structure, its geometric size and the material properties for each layer are determined by cell images and measurement results of indentation experiments, respectively. The model shows high simulation accuracy with the detailed response to the external force stimuli. Although the FEM model performed fairly well in simulating the living cells, this method requires lots of manual measurements to obtain the model size and material properties since those parameters vary according to the cell types. To address this issue, Chapter 3-5 migrate the focus inside the cell–cellular cytoskeleton, which is a kind of interlinking protein filaments present in the cytoplasm of all cells. In Chapter 3, we propose an image recognition-based actin cytoskeleton quantification (IRAQ) approach. IRAQ aims to quantify both the actin cytoskeleton orientation and quantity simultaneously through pre-designed parameters. Since both actin cytoskeleton and microtubules can reorganize their network structures to control the cellular mechanical properties, to get in-depth understandings of the cellular adaptive response to the external stimuli, we use the extended IRAQ model to report the quantitative investigation on the effects of the actin cytoskeleton and micro- tubules in affecting both the elasticity and poroelasticity in Chapter 4. However, the models proposed in Chapter 3 and 4 extracts the manually selected features for the cytoskeleton quantification. Those features may be too simple to fully deliver the information contained in the cytoskeleton structure. Therefore, an automated analytical model is needed for further studies. We address this issue in Chapter 5. In Chapter 5, a machine learning model to recognize the cellular actin cytoskeleton morphology was built using the graph to vector embedding technique together with a fully connected neural network. The test loss and accuracy were obtained at the end to show the model performance in extracting features of actin cytoskeleton from the fluorescent images. The proposed model was also compared to a general CNN and three commonly used transfer learning models (GoogleNet, Xception, and VGG16) to prove its ability in avoiding overfitting and keeping the model generalization.