M2+PhD: Direct inference of cell rearrangements in collective cell migration
J.-F. Rupprecht team, Centre de Physique Théorique, Marseille Luminy.
Objectives Currently, the detection of cellular rearrangements in biological tissues relies on an extremely difficult step: the segmentation and tracking of cell-cell interfaces. In this internship, we propose bypassing this step and inferring the positions and directions of the rearrangement directly from time-lapse images. We will readapt a method that our team initiated to locate cell divisions (Karnat et al. bioRxiv 2024). This internship serves as the starting point to PhD in Physics, aimed at understanding the phase separation in tissues according to their fluidity, with applications to the symmetry breaking in 3D gastruloids – a model system for embryogenesis.
The team Jean-François Rupprecht is a theoretician combining continuum (active nematics, Prasad et al. Science 2024) and cell-based (vertex, Lin et al. PRL 2023 & Sonam Nat. Phys. 2023) models. We closely collaborate with Sham Tlili, physicist, leading model, analysis & experiments for gastruloids (Gsell et al., Nat. Phys. 2024).
Image Stage M2 team Rupprecht
Context Cellular rearrangement is the process by which cells change their position and organisation within a tissue, akin to bubbles in a foam that reorganise to relieve mechanical stresses, see Fig. A. Rearrangements are represented by a tensorial quantity (Graner et al. 2008). Maps of these tensors have been obtained in the context of the development of the thorax and wing of Drosophila in the pupa stage (Guirao Science 2016). However, the ideal conditions of the Drosophila pupa are rarely met: in most cases, the real-time imaging of biological tissues is limited by constraints such as phototoxicity in fluorescence microscopy, which restricts temporal and spatial resolution. Constraints on cell segmentation are even greater in most cases, as in 2D MDCK in Fig. B, or 3D.
In this M2 internship, we propose to directly detect the positions and orientations of rearrangements using dedicated networks (see Fig 1A), based on the preliminary results obtained during the PhD of our student, Marc Karnat (see reference below). To bypass the tedious task of building a training dataset through manual annotation in experimental movies, we recently adapted a generative adversarial network (GAN) to transfer the style of experimental images to the binary tissue masks obtained by vertex simulations (Fig B). These simulations are then used to build a training dataset (Fig. C) The preliminary results of this training method based solely on numerical simulations are very promising: many rearrangements are correctly detected in our experimental data of collective migration of MDCK cells around obstacles.
Ph.D. subject Analysis of the rearrangement tensor enables the quantification of tissue fluidity. We propose a Ph.D. subject on the physics of cell sorting based on fluidity modulations, which builds upon our recent success in modelling tissues according to their viscosity levels (Fu et al. PNAS 2024).
We will perform both analytical and simulations work. Beyond studying a beautiful system at the interface between several fields of physics (active matter, granular systems, stat. phys), I believe this M2 internship & Ph. D. subject will also empower the student with a set of deep learning methods that are instrumental in a large range of career paths - from IA-based industry jobs to academia.
Possible PhD funding Yes, I have fundings available. Starting date ranging from Jan. 2025 onwards.
Location at Marseille Luminy,
arguably the most beautiful campus in the world, right in the middle of the Calanques National Parc.
Contact jean-francois.RUPPRECHT@univ-amu.fr ;
rupprecht.jf@gmail.com
Former student contact:
marc.karnat08@gmail.com
