Quantifying bacterial order in live three-dimensional Myxococcus xanthus colonies

Host laboratory and collaborators

Jean-François Rupprecht (CPT) /

Tâm Mignot (LCB) /


Myxococcus xanthus are rod-shaped bacteria that pack along biofilm structures, which can take a large variability of shapes. Motivated by the understanding of the ecology of bacterial colonies, we aim to understand the relation between the bacteria packing order and the overall shape of the biofilm. Yet tools are lacking to quantify and track bacterial cells in complex three-dimensional environments. In this project, we propose to design an advanced image analysis framework to measure the field of local bacterial packing order using deep learning methods. In particular, we will leverage recent development in style-transfer methods (diffusion models) to simulate experimental images based on numerical simulations.


Bacterial Ecology, Active Matter Mechanics, Deep Learning Methods, Generative Adversarial Networks


-       Design a new pipeline for quantifying the bacteria packing order
-       Test & optimize the pipeline using realistic simulations
-       Develop a coarse-grained active nematic theory based on simulations
-       Relate the bacterial nematic order to the overall biofilm shape and properties

Proposed approach (experimental / theoretical / computational)

In recent years, several methods emerged to analyze the organization of dense bacterial communities over flat, 2D substrates [Ref. 5]. However, these solutions are inefficient for studying most bacterial communities which organize in 3D. To address this problem, we propose to use Myxococcus fruiting bodies as a model system, which have the advantage of being well organized assemblies. We will image these structures by confocal microscopy. To reconstruct the bacteria order field, we will expand the capabilities of our in-house, supervised-deep-learning-based method [Refs 5,9]. To enhance the size of our training dataset, we propose to use simulations, following an established approach in the field. Following two recent works [Refs. 1,2], we propose to use CycleGAN (Generative Adversarial Networks) to generate realistic bacterial datasets (with spatial and temporal continuity) directly from simulations of ellipses interacting through Lennard-Jones potential, using the popular ya||a code [Ref. 4]. These simulations will also be helpful to test continuum active nematics models [Ref. 8,10]. Once our approach proves effective on the Myxococcus system, we will expand it to other ecologically-relevant bacterial communities, e.g. alcanivorax borkumensis [Ref. 8].


The project is fundamentally interdisciplinary, merging:
-       IA-based image analysis methods
-       Physical modeling (active nematics)
-       Bacterial cell biology
-       Bacterial ecology


Expected profile

We are looking for a student who enjoys coding and has had some hands-on machine learning; interest in biological questions and image analysis is also key.

Is this project the continuation of an existing project or an entirely new one? In the case of an existing project, please explain the links between the two projects

This project is an entirely new one.

2 to 5 references related to the project

1. Liu, Q. et al. ASIST: Annotation-free Synthetic Instance Segmentation and Tracking by Adversarial Simulations. (2021).
2. Zhang, J. et al. BCM3D 2.0: Accurate segmentation of single bacterial cells in dense biofilms using computationally generated intermediate image representations. bioRxiv (2022)
3. Wang, J. et al. 3D GAN image synthesis and dataset quality assessment for bacterial biofilm. Bioinformatics 38, 4598–4604 (2022).
4. Germann, P., Marin-Riera, M. & Sharpe, J. ya||a: GPU-Powered Spheroid Models for Mesenchyme and Epithelium. Cell Syst 8, 261-266.e3 (2019).

3 main publications from each PI over the last 5 years

Tâm Mignot

5.     A Tad-like apparatus is required for contact-dependent prey killing in predatory social bacteria. Seef S, Herrou J, de Boissier P, My L, Brasseur G, Robert D, Jain R, Mercier R, Cascales E, Habermann BH, Mignot T. eLife (2021)
6.     Misic, a general deep learning-based method for the high-throughput cell segmentation of complex bacterial communities. Panigrahi S, Murat D, Le Gall A, Martineau E, Goldlust K, Fiche JB, Rombouts S, Nöllmann M, Espinosa L, Mignot T. eLife. (2021)
7. The polar Ras-like GTPase MglA activates type IV pilus via SgmX to enable twitching motility in Myxococcus xanthus. Mercier R, Bautista S, Delannoy M, Gibert M, Guiseppi A, Herrou J, Mauriello EMF, Mignot T. PNAS (2020)


Jean-François Rupprecht

8. Alcanivorax borkumensis Biofilms Enhance Oil Degradation By Interfacial Tubulation
M. Prasad, N. Obana, S.-Z. Lin, K. Sakai, C. Blanch-Mercader, J. Prost, N. Nomura, J.-F. Rupprecht*, J. Fattaccioli*, A. S. Utada*, bioRxiv (2022)
9. Mechanical stress driven by rigidity sensing governs epithelial stability
S. Sonam, L. Balasubramaniam, S.-Z. Lin, ..., J.-F. Rupprecht*, B. Ladoux*, Nature Physics (2022)
10. Active nematic flows on curved surfaces S. Bell, S.-Z. Lin, J.-F. Rupprecht*, J. Prost*, Physical Review Letters (2022)