The cell cycle is fundamental to life, governing the growth, replication, and maintenance of all living organisms by ensuring accurate cell division and genetic integrity across generations. Yet despite its central role, the dynamics of the cell cycle and its interplay with other biological processes remain obscured within the complexity of living systems. In particular, how cell-cycle regulation interacts with early mammalian development and influences processes such as cell fate decisions remains incompletely understood. This thesis addresses these questions from complementary angles, examining the biological consequences of perturbing the cell cycle in early mammalian development and advancing methods for the quantitative analysis of cell-cycle dynamics.
The first part of this thesis investigates the consequences of cell cycle inhibition during early mammalian development. Using mouse gastruloids, three-dimensional models that recapitulate key features of gastrulation, we show that even under near-complete growth arrest, major developmental events such as symmetry breaking, elongation, and germ layer formation still occur. Nevertheless, consistent shifts in cell type proportions emerged, reflecting delayed differentiation and lineage-specific shifts in population balance. These findings indicate that while the developmental processes can proceed in the absence of a normal cell cycle, it remains finely tuned by its dynamics.
The importance of accurately characterising these dynamics motivated the development of computational approaches for their quantitative analysis. Accordingly, the second part of this thesis focuses on inferring cell cycle phase directly from label-free brightfield imaging. To achieve this, a deep learning framework was developed and trained on a newly generated dataset of over a million time-lapse images of cycling cells to predict continuous Fucci signals. Causal and transformer-based models accurately reconstructed cell cycle architecture, correctly identifying biological events such as G1/S transitions. This approach enables the monitoring of proliferation dynamics without the need for engineered reporter systems, demonstrating the potential of label-free approaches to study cell cycle dynamics.
As sequencing technologies increasingly shape modern biology, single-cell transcriptomic data open new possibilities for resolving cell cycle dynamics within complex populations. Therefore, the final part of this thesis introduces VeloCycle, a Bayesian framework that models cell cycle dynamics from single-cell RNA sequencing data, a project led by Alex Lederer in the La Manno lab. By leveraging unspliced and spliced transcript information from periodic genes, VeloCycle captures differences in cycle speed across cell populations and perturbations, offering a statistically consistent framework for RNA velocity inference.
Together, this work shows that mammalian development can proceed under cell cycle inhibition, though with al
