Statistical Texture-Based Mapping of Cell Differentiation Under Microfluidic Flow

Timelapse microscopy imaging allows for long term monitoring of biological processes however a major bottleneck in assesing experimental outcome is the need for an automated analysis framework to extract essential statistics. In this study, we use Gabor texture descriptors to generate a high dimensional feature space which is analysed with PCA to provide unsupervised characterisation of texture differences between pairs of images. We apply this technique to differentiation of carcinoma cells in the presence of retinoic acid (RA) and show that differentiation outcome can be predicted directly from texture information. A microfluidic environment is used to accurately control the timing of RA signal over five days in culture. Results provide insight into the dynamics of cell response to differentiation signals over time. 1 Scientific Background Embryonic stem cells have the ability to generate all cells in the adult body through differentiation, a process by which cells acquire highly specialised function and morphology. In the developing embryo, cell differentiation undergoes a complex spatial patterning with very precise timings difficult to replicate in vitro. As a result, differentiation of cells in culture is inefficient and not well understood. Recent studies indicate that cell response to external factors can be amplified by pulsing of signals as opposed to constant exposure [1, 2]. Two main technologies can be used to enable control of input signal properties, fast solution switching in microfluidic environment [1] and photoactivatable systems [2]. Microfluidics have the advantage of delivering multiple inputs and controlling other culture parameters such as shear stress [3]. We investigated the differentiation of NTERA2 human embryonic carcinoma cells in controlled microfluidic flow. When exposed to retinoic acid, NTERA2 are known to differentiate towards a non-neural fate identified by surface marker ME311 and neural cells detected after 3-4 weeks [4] , however expression at early stages in differentiation is not fully understood. In this study, we combine timelapse imaging and unsupervised characterisation of texture to describe the timeline of early differentiation. 2 Materials and Methods Texture feature set using Gabor energy. Texture descriptors can be used to decompose intensity values in an image I(u, v) into sub-bands with preffered orientation and sparial frequency, revealing hidden spatial information. The Gabor filter [5] implies Proceedings of CIBB 2016 2 kernel convolution with a Gaussian kernel g: gλ,σ,γ,θ,φ(x, y) = e − x′+γy′ 2σ2 cos (