The differentiation of human stem cells is a complex and highly orchestrated spatiotemporal process, where cell morphology, proliferation and cell cycle properties likely play essential roles in many cell fate decisions. Recent independent advances in next generation sequencing (NGS) and live microscopy imaging technologies have greatly pushed forward stem cell differentiation research and have highlighted the need to combine spatial-temporal information derived from microscopy with transcriptional information accessible by NGS, as a way to better understand the functional links between spatiotemporal, morphological and cell cycle dynamics with cell fate transitions. In this thesis we developed novel technological pipelines aiming to address that need. In a first part of the thesis, we prototyped a method termed positional photoprinting (POPH) that could in principle allow to record the spatial localization of single-cells at the end of timelapse imaging experiments as a way to enable imaging-derived information to be linked to transcriptional information acquired post imaging, by leveraging photoactivatable protein (PA-protein) technologies. In parallel we developed a conceptual computational tool, Selection-seq, allowing to link imaged cell phenotype to transcriptome from PA-protein expressing cell lines in a Machine Learning (ML) style. In a second part of the thesis, we analyzed published single-cell RNA sequencing (RNA-seq) data to screen for gene candidates that might constitute a link between cell cycle and early neuronal differentiation control, including genes involved in cell cycle regulation (CDKs and Cyclins), axon generation and microtubule assembly regulation (KIF families), cell pluripotency (AURKA) and neurogenesis (MAP2, SOX2) , and later used a Fluorescence Ubiquitin Cell Cycle Indicator (FUCCI) expressing human Neural Stem Cell (hNSC) line processed both by live imaging and gene expression profiling to explore the possible temporal linkage between cell cycle dynamics and neuronal differentiation in hNSCs, using a graph-theory based workflow we established that allows to make a integrative analysis of gene interactions using both experimental and meta-analysis data. We found a high correlation among our gene list of interest in terms of both expression profile and known putative protein interactions, and found that the genes can be grouped into 2 distinct modules with opposite expression pattern during neuronal differentiation. In a third part of the thesis, we demonstrate how Machine Learning and multi-focus image fusion technologies can be applied to establish an image analysis workflow of high-throughput live imaging data that calibrates vignetting, integrates information from multiple z-plane focal layers, quantifies neuronal process images challenging to conventional analysis, and quality-controls cell lineage data obtained from cell tracking analysis. By quantitatively analyzing live imaging data obtained from FUCCI and microtubule-labelled hNSCs differentiating into neurons, we found a global correspondence between cell cycle phase and neurite outgrowth, spatially and temporally. Altogether the novel technical insights and technological pipelines and solutions presented here could potentiate integration of ‘live’ microscopy and gene expression derived phenomics information across many other cellular control processes, cellular differentiation paradigms and cell therapy applications.
Quantitative and Machine Learning-enhanced microscopy approaches allowing to investigate morphological, proliferative and transcriptional changes during early human stem cell differentiation
Shi, Y. (Author). 20 Jun 2023
Student thesis: Doctoral Thesis › Doctor of Philosophy (PhD)