|M.Sc Student||Megides Or|
|Subject||Morphokinetic Single-Cell Analysis and Machine Learning as a|
Tool to Characterize Breast Cancer Cell Motility
and Response to Therapy
|Department||Department of Biomedical Engineering||Supervisors||Professor Emeritus Eitan Kimmel|
|Professor Ilan Tsarfaty|
Breast cancer is the most frequently diagnosed cancer in women. Triple-negative breast cancer (TNBC) is negative for estrogen and progesterone receptors, excess HER2, higher metastatic potential, and a poorer prognosis. Cancer cell motility, invasion, and energy metabolism are fundamental steps in metastasis. Combined precision-targeted therapy and chemotherapy offer a novel anti-metastatic treatment. MET, a tyrosine kinase receptor and its ligand, Hepatocyte Growth Factor/Scatter Factor (HGF/SF), induce specific signal transduction in tumor cells, leading to cell motility and metastasis. MET's induced tumorigenesis and metastatic processes make it an ideal target for anti-cancer therapy.
In this work, we characterized different motility patterns of three cell types: estrogen receptor-positive (EP), TNBC, and Madin-Darby Canine Kidney (MDCK). We also described the effect of MET inhibition, chemotherapy, low-intensity ultrasound (LIUS), and their combined impact on breast cancer motility. Time-lapse fluorescence images of EP, TNBC, and MDCK cells were subjected to commercial packages for segmentation and tracking - extracting morphokinetic (MK) information at a single-cell level.
To better understand cell motility patterns, we developed the infrastructure, Tool for Analysis of Single-Cell (TASC), based on unsupervised machine learning methods to demonstrate a high-dimensional feature set at a single-cell resolution. We compared our experimental MK results to simulations of three modified classical physical models: Lévy flight (LF), fractal Brownian motion, and random walk (RW) with or without a back-propagating wave (BPW).
TASC analysis demonstrated a fundamental difference between EP and TNBC cells: both cell types were constructed from two subpopulations harboring similar MK characteristics. TNBC cells presented an additional subpopulation characterized by 1) dominantly increasing cumulative kinetics values (MSD, displacement2), 2) highest temporal wave values (velocity starting-time, velocity maximum-height), and 3) a decrease in non-cumulative kinetics values. The combined treatment of TNBC cells with MET inhibition and chemotherapy eliminates this high-motility group.
Cell motility models and motion simulations were compared using Wasserstein’s analysis. Untreated TNBC cells showed high similarities to RW with BPW simulations, while MET-activated TNBC cells transformed into LF without BPW simulations, indicating more aggressive behavior.
In comparison to MK features, we used the Heteromotility tool, which provides motility features using a temporal moving average analysis. Both feature sets revealed one high-dynamic subpopulation of TNBC cells. The Heteromotility presented a better separation between the combined treatment of MET inhibition and chemotherapy with untreated TNBC cells. However, morphological parameters from the MK feature set were advantageous in clustering the two 'slow' subpopulations.
TASC analysis revealed previously undefined high motility subpopulation of TNBC that was eradicated by combining MET inhibition and chemotherapy. Moreover, TASC analysis demonstrated the morphology alteration in MET activated and non-activated MDCK cells treated with LIUS.
The primary benefit of TASC is the ability to cluster and characterize similar behaviors across different motility models while still detecting subtle changes within each one. MK analysis combined with TASC can lead to discovering novel targets for precise therapy. Moreover, this infrastructure may determine patient tumor susceptibility to chemotherapy and biological treatments and operate as an analytical tool for personalized targeted therapy.