Dallish

Equipe(s)
Agence de moyen
Etat
Titre du projet
Data Assimilation and Lattice Light Sheet Imaging for endocytosis/exocytosis pathway modeling in the whole cell
Nom de l'appel d'offre
ANR blanche
DĂ©fi/axe ANR
éfi : Société de l'information et de la communication Axe : Données, Connaissances , Données massives
Coordinateur.trice
C. Kervrann (INRIA Rennes)
Participants de MaIAGE
A. Trubuil, S. Laparthe, B. Laroche
Partenaires (hors MaIAGE)
INRIA-SERPICO, Institut Curie
Année de démarrage - Année de fin de projet
2016-2019
Date de fin du projet
Résumé
Context: Fluorescence imaging and microscopy has a prominent role in life science and medical research. It consists of detecting specific cellular and intracellular objects of interest at the diffraction limit (200 nm), using wide field as well as confocal microscopy, after tagging them with genetically engineered proteins that emit fluorescence. In this context, the past decade has witnessed a tremendous interest in the concept of Super-Resolution Microscopy (SR-M) and 3D fluorescence Light Sheet Microscopy (LS-M). One of the main reasons explaining this enthusiasm lies in the discovery of methods to break the diffraction limit established in optics several decades ago. Since, variants of SR-M (awarded by Nobel Prize in 2014) and LS-M have been investigated in an increasing number of biological studies. The so-called Lattice Light Sheet Microscopy (LLS-M) represents at present the novel generation of 3D fluorescence microscopes dedicated to single cell analysis, generating extraordinarily sharp, 3D images and videos. However, the usual conventional image processing algorithms developed for fluorescence microscopy are likely to fail to process the deluge of voxels generated by LLS-M instruments. The DALLISH project aims at improving the core of 3D image processing and quantification methods to face this computational challenge. The proposed methods will be used to decipher mechanisms involved in protein transport observed in LLS-M experiments.

Reconstruction of very huge 3D images: Inverse problems and image processing may be intractable with LLS-M because we are facing very large temporal series of volumes (200–1000 images per second for one 3D stack) acquired for several hours. These volume series represent several hundreds of Gigabytes if the LLS-M is used to study the 4D interplay of several proteins in the same cell at the same time. Accordingly, the usual methods need to be extended and novel strategies have to be found since computation is extremely heavy in LLS-M experiments. First, we will address the problem of image reconstruction (including deskewing) and deconvolution, which are the core and mainstream parts of LLS-M. All 3D data sets have to be deskewed to account for the 31.8° angle of the detection objective. Deconvolution of 3D images reduces blurring from out-of-focus light and enables quantitative analyses, but existing software for deconvolution is slow and expensive. We are aiming for parallelized methods that reconstruct and deconvolve 3D images faster than conventional software (few seconds per image) and run on a low-cost graphics processor board (GPU).

Image analysis of huge 4D images: Several image analysis methods need to be developed to quantify intracellular trafficking. First, we will study the problem of detection and segmentation of individual molecules/spots and extend the state-of-the art methods developed in 2D imaging to 3D LLS-M. Second, we will adapt the usual optical flow and tracking methods in fluorescence microscopy to meet the requirements in 4D LLS-M imaging. We will focus on the recent optical flow methods that combine local and global approaches and use the concept of estimator aggregation. An additional promising paradigm based on temporally varying object counting on multi-scale 3D spatial grids will be examined in the project. These segmentation and motion analysis methods and algorithms will be used to track endocytic and exocytic building blocks in time and 3D over the entire volume of a cell, such as to be able to follow individual transport vesicles and molecules propelled by motor activity, from the cell surface into the cell (endocytosis) or from endosomes to the cell surface (exocytosis).

Quantitative microscopy with biophysical models: Given image motion-based features and tracklets (small trajectories), we will investigate quantification methods to represent interactions between molecules and trafficking around three lines of research. First, we will study 3D space-time global and local object-based co-localization methods to quantify interactions between molecular species. Second, the dynamics of trajectories will be classified into three categories: Brownian motion, confined diffusion, directed diffusion. Therefore, statistical tests will be developed to replace conventional methods (e.g. MSD), which cannot reliably discriminate these three motion types. Moreover, the detection of regime changes occurring along trajectories will be modeled by generative models and analyzed in the Bayesian framework. Finally, given N tracks associated to N molecular species, stochastic models representing interactions between the molecular species will be also studied in the Bayesian statistical framework. Third, we will investigate approaches to estimate molecular mobility and active transport from the computed trajectories or optical flow descriptors. We will investigate the concept of super-resolution to provide spatially high-resolved maps of diffusion and active transport parameters based on stochastic models. An additional approach will be based on sparse image representation combined with biophysics to localize molecules at higher-resolution than the LLS-M resolution.

Endocytosis and exocytosis pathways analysis: Understanding the molecular and cellular mechanisms underlying membrane traffic pathways is central in fundamental cell biology and crucial to the treatment and cure of human disease: various human diseases caused by changes in cellular homeostasis arise through a single gene mutation; pathogenic agents such as viruses, bacteria, or parasites have evolved mechanisms to corrupt the host cell response to infection. Here, we propose to combine LLS-M, quantitative bioimaging, individual-based modeling and cell biology to assess the functional properties of molecular complexes and elucidate the role of key molecules involved in the endocytosis and exocytosis pathways. To this end, the algorithms, gathered in well-defined workflows, will be exploited to analyze experimental data composed of several series of proteins expressed at endogenous levels after genome editing. The framework will be quite flexible to adapt to a large range of biological studies and 4D microscopy modalities.
Année de soumission
2015