(229at) Image-Based Predictive Computer Model of Mesenchymal Stem Cell Migration in Physiologically-Mimicking 2D Microenvironments
Cell migration is important to a number of biological phenomena ranging from embryonic morphogenesis and wound repair, to cancer invasion and immune response. In particular, the study of stem-cell migratory behavior is of great interest to tissue engineering applications, because the cells are prone to uncontrolled migration within the 3D scaffold-pore environments. This in turn leads to lack of quality-control over the resulting tissue architecture, and ultimately gives rise to product variability. However, developing an understanding of stem-cell migration in complex environments is challenging. This is because tissue engineering studies are frequently done by seeding cells on 3D scaffolds in which real-time non-invasive observation is not possible. As a result, most of the analysis is done by performing bulk differentiation and proliferation assays on crushed samples. In other words, only single time point data corresponding to the end of the experiment is available, and it is not single-cell level. Moreover, limited biological understanding of the underling migration mechanism renders bottom-up modeling impractical. Therefore, in order to provide insight into stem cell migration in the tissue engineering scaffolds, we approach the problem in a top-down manner: a machine-learning algorithm is trained on time-lapse microscopy images collected from microfluidic devices meant to mimic the engineered tissue scaffold pores.
Since mesenchymal stem cells (MSCs) are typically used as the starting cell type for most tissue engineering applications, these are selected as the primary cell type of choice for the study. The migration experiments are performed in 2D microfluidic â??mazeâ? platforms, as they are most representative of the complexity of the scaffold pore geometries. Namely, these devices contain multitude of longer paths, shorter paths, and dead ends in which cells are allowed to make a range of migration decisions. This allows us to study a larger variety of cell migration decisions in parallel. The migration of MSCs is chemotactically induced by the diffusion-based concentration gradient of platelet-derived growth factor-BB (PDGF-BB) and fetal bovine serum (FBS). The whole cell migration process in the microfluidic device is monitored and imaged in real time using an inverted microscope (IX83, Olympus) installed with a custom incubation chamber. The stability of the chemical gradient is tracked via fluorescent 20 kDa Dextran-FITC, while the initial concentration gradient of PDGF-BB and FBS is computationally verified via COMSOL modeling based on their molecular weights. Different maze dimensions as well as different chemotactic profiles affecting the migration of cells are also investigated.
At each time point a high resolution phase-contrast image is captured by the robotic microscope. The stem cells and their organelles are segmented from the image, without fluorescent markers, using computer vision (a highly-efficient Maximally Stable Extremal Region detector). The imaged device geometry and cell locations are exported into a mass-transport solver, which calculates the chemo-attractantâ??s distribution within the microfluidic channels. The solved concentration field is exported back and overlaid on the imaged geometry. Each pixel is then sampled and descriptive statics (e.g., chemotactic gradient value, distance away from the device wall, distance away from previous cell position, etc.) are calculated for use as features. A machine-learning classifier is trained on the experimental data to classify each pixel as cell or not-cell based on these features. The end goal is the ability use the classifier for prediction of likely cell migration paths in maze geometries not previously encountered by the machine. Ultimately, such a model could be extended to 3D environments, in order to shed insight into the tissue engineering scaffold micro-pores, which cannot be otherwise imaged during culturing. Knowledge of cell behavior within the pores would be beneficial for designing novel culturing methods and tissue engineering approaches, which are currently done largely by trial-and-error.