ED Sciences de la Vie et de la Santé
In depth Single Molecule Localization Microscopy within 3D multicellular models using Adaptive Optics and PSF engineering
by Laetitia BETTAREL (Institut Interdisciplinaire de Neurosciences)
The defense will take place at 14h00 - Amphithéatre Centre Broca Centre Broca Nouvelle Aquitaine IINS - UMR 5297 - Bât. Neurocampus 146 rue Léo Saignat
in front of the jury composed of
- Rémi GALLAND - Chargé de recherche - Université de Bordeaux - Directeur de these
- Lydia DANGLOT - Chargée de recherche - Institut de Psychiatrie et de Neurosciences de Paris (IPNP) - INSERM - Rapporteur
- Alexandra FRAGOLA - Professeure des universités - Institut des Sciences Moleculaires d'Orsay (ISMO) - Rapporteur
- Laurent COGNET - Directeur de recherche - Laboratoire Photonique, Numérique et Nanosciences (LP2N) - Examinateur
Assessing protein organization and dynamics in their native cellular context provides key insights into the molecular mechanisms that govern cell function. Super-resolution microscopy has been a major breakthrough in this regard, driving major discoveries in cell, developmental and neuro-biology. Amongst these techniques, Single Molecule Localization Microscopy (SMLM) enables locating, tracking and counting biomolecules in their cellular environment with nanoscale resolution. However, conventional SMLM imaging is restricted by its shallow penetration depth, precluding many biological events to be investigated. Performing volumetric SMLM deep within complex multicellular samples therefore poses several challenges: achieving efficient optical sectioning with high photon collection capabilities, and correcting the optical aberrations introduced both by the optical system and the sample, which blur the single molecule signals and compromise localization precision and accuracy. To address these challenges, we developed in the team a specific light-sheet architecture, named soSPIM, which enables in-depth single-molecule imaging and supports the culture and observation of complex 3D cellular models. In parallel, Adaptive Optics (AO) has emerged as a powerful solution to correct system- and sample-induced aberrations and thereby improve image quality in-depth. Recently, we combined soSPIM with AO to achieve volumetric 3D SMLM imaging at the whole cell scale. Yet, this implementation still relies on fiducial markers located close to the sample, which do not allow the effective correction of sample-induced aberrations, that become especially significant within multicellular systems. In addition, it uses conventional 3D localization approaches, that are non-optimal for fast and accurate in-depth single molecule localization. In this context, my PhD work focused on developing methodological solutions to extend the applicability of the AO-soSPIM imaging platform for in depth SMLM in complex 3D samples. First, I developed a fully custom Python-based sensorless AO correction algorithm allowing complete control over all parameters of the correction loop, including the integration of user-defined image quality metric specifically tailored to the imaging modality and sample type. Building on this, I established a systematic framework to assess fiducial-free image-based metrics and identify those most sensitive and robust under in-depth SMLM experimental conditions. Together, these developments provide a versatile and reliable foundation for restoring diffraction-limited performance in photon-limited SMLM acquisitions. Second, I investigated deep learning-based single molecule localization frameworks that exploit data-driven PSF models to enhance both localization robustness and imaging speed, offering a promising alternative to conventional Gaussian fitting in dense or challenging 3D SMLM datasets. I also explored experimental PSF modeling strategies to better account for residual aberrations and complex PSF deformations in depth. These approaches, which capture PSF shapes beyond the Gaussian approximation, aims to improve localization precision and accuracy under aberrated conditions. Altogether, these methodological developments establish a robust pipeline for aberration-corrected, high-resolution 3D SMLM within complex biological 3D samples. By enabling reliable volumetric SMLM imaging beyond the coverslip, this work broadens the scope of super-resolution imaging toward physiologically relevant 3D models such as spheroids and organoids.
Identification of leukemic stem cells in a novel transgenic model of chronic myeloid leukemia
by Yasmine TAVEAU (BoRdeaux Institute of onCology)
The defense will take place at 14h00 - Salle de Conférence Centre d'Appui à la Recherche et de Formation (CARF) 146 rue Léo Saignat 33000 Bordeaux
in front of the jury composed of
- Béatrice TURCQ - Chargée de recherche - Université de Bordeaux - Directeur de these
- Bastien GERBY - Directeur de recherche - Centre de Recherches en Cancérologie de Toulouse (CRCT) - Rapporteur
- Franck-Emmanuel NICOLINI - Praticien hospitalier - Centre Léon Bérard - Rapporteur
- Stéphane PROST - Directeur de recherche - CEA Fontenay-aux-Roses - Examinateur
- Katia BONIFACE - Professeur - Université de Bordeaux - Examinateur
Chronic myeloid leukemia (CML) is a myeloproliferative disorder that is initiated when a hematopoietic stem cell (HSC) acquires the BCR::ABL1 fusion oncogene following a chromosomal translocation between chromosomes 9 and 22. This oncogene encodes the BCR::ABL1 protein, a constitutively active tyrosine kinase, responsible for pathological proliferation of the granulocyte lineage compartment. CML begins with a chromic phase that can last several years, and if left untreated, the disease progresses to an accelerated phase and then to blast crisis. The development of tyrosine kinase inhibitors (TKI) has revolutionized the treatment of the disease, providing targeted therapy against cells expressing BCR::ABL1. The spectacular results of this new therapeutic class have even led to the suggestion of treatment discontinuation attempts in patients with very good molecular response. Unfortunately, more than 50% of patients relapse upon treatment discontinuation due to the persistence of leukemic stem cells (LSC), which does not yet allow the use of the term "cure" in this disease. The identification and characterization of the LSC, as well as the mechanisms of their persistence, remain understudied due to a lack of relevant models. The objective is to develop a preclinical model of the chronic phase of CML to characterize the LSC responsible for relapses. This new mouse model, Pdzk1ip1-CreERTg/+ TRE-BCR-ABL1Tg/+ R26rtTA-GFP/Tom (hereinafter Cre+ BA+), allows the inducible expression of a BCR::ABL1 transgene in a fraction of HSC. The differential expression of fluorescent markers allows to track normal hematopoietic cells versus BCR::ABL1+ cells. Unlike existing models, this model incorporates the key features of human CML, including a long chronic phase, allowing the study of the LSC. To identify disease-initiating cells, which are likely responsible for relapses upon treatment discontinuation, functional transplantation tests were performed on different populations of Tom+ GFP+ leukemic cells derived from either the bone marrow or spleen of Cre+ BA+ animals. Only cells with an HSC phenotype (Lin- Sca1+ cKit+ CD150+ CD48-) were capable of initiating the disease in recipient mice. Concurrently, induction of BCR::ABL1 expression in Cre+ BA+ animals led to the onset of CML symptoms. Animals were treated with daily administration of dasatinib, a second-generation TKI. After 30 to 40 days of treatment, a decrease in Tom+ GFP+ granulocytes, normalization of platelets and white blood cells, and a decrease in BCR::ABL1 transcripts were observed. After 50 days, treatment was stopped and in the majority of mice, disease symptoms reappeared, demonstrating the persistence of cells capable of reinitiating the disease and the relevance of this model for the study of cells responsible for relapses during treatment withdrawal attempts in CML. At the end of treatment, residual LSCs were still present in the bone marrow, evidence of disease persistence. To explore the regulatory mechanisms of persistent cells, single-cell transcriptomics on Tom+ GFP+ hematopoietic stem cells (Lin-cKit+ Tom+ GFP+) was performed and alterations in gene expression and specific biological pathways, as well as marker genes for these cells were identified. The identification and characterization of leukemic stem cells (LSCs) could reveal new therapeutic strategies that could lead to a truly curative solution for CML. The development of this preclinical mouse model, which reproduces the initiation and chronic phase of the disease, represents a major asset for the study of these cells and the development of innovative treatments.
AGE DEPENDENT RESTRICTION OF ADENOVIRUS INFECTION IN PRIMARY BRONCHIAL EPITHELIA
by Anastasiia TYMCHENKO (Microbiologie fondamentale et Pathogénicité)
The defense will take place at 14h00 - Module 1.1 146 Rue Léo Saignat, bât CROUS 1er étage, 33000, Bordeaux
in front of the jury composed of
- Goujon CAROLINE - Directrice de recherche - Institut de Recherche en Infectiologie de Montpellier - Rapporteur
- Ehrhardt ANJA - Full professor - Witten/Herdecke University - Rapporteur
- Johan GARAUDE - Associate Professor - University of Bordeaux - Examinateur
- Julie DéCHANET-MERVILLE - Directrice de recherche - University of Bordeaux - Examinateur
Adenoviruses are respiratory viruses that primarily infect the upper airways and usually cause mild symptoms such as the common cold, largely controlled by life-long immunity established after childhood infections. However, certain genotypes are linked to more severe respiratory disease accompanied by proinflammatory cytokine release. This thesis aims to understand how adenoviruses spread in bronchial epithelia (BE), the intrinsic mechanisms regulating this process, and how viral infections can induce cross-protection against other respiratory viruses. We established a model of primary BE grown at the air–liquid interface from lung brushes of healthy adult and pediatric donors. Using C-type adenoviruses (CAR receptor) and B-type adenoviruses (DSG2 receptor), we tracked viral spread with genetically tagged strains and clinical isolates through live imaging and confocal microscopy. Epithelial integrity and apical/basolateral viral release were monitored to characterize infection dynamics. To define epithelial responses, we analyzed transcriptional changes at peak infection using scRNA-seq. Adenovirus infection progressed from local foci that developed into macroscopic lesions, coinciding with a decline in transepithelial electrical resistance, indicating loss of integrity. Spread occurred first apically, then basolaterally after epithelial damage. No major differences were found between BEs from adult and pediatric donors. Our data suggest a model where local foci gradually compromise epithelial integrity, allowing trans-epithelial spread. At the transcriptional level, infection induced a distinct cell subpopulation with upregulated FAM111B, a putative restriction factor. Single-cell RNA-seq further revealed enhanced interferon signaling and strong induction of interferon-stimulated genes (ISGs). We therefore investigated whether virus-induced interferon responses confer cross-protection against other respiratory viruses. BEs from both donor groups were infected with three common respiratory viruses differing in genome type, replication strategy, and innate immune kinetics: adenovirus, rhinovirus A (RVA), and rhinovirus C (RVC). We compared replication dynamics, innate immune responses, and infection morphology, and profiled transcriptional responses to RVA and RVC at peak infection using scRNA-seq. To test cross-protection, BEs pre-infected with each virus were challenged with GFP-expressing RVA, and infection was quantified after 24 hours.All three viruses induced cross-protection, though with distinct kinetics. Adenovirus showed delayed infection dynamics, leading to correspondingly delayed immune responses. Bulk RNA-seq of cross-protected vs. non-protected BEs identified a distinct ISG signature associated with epithelial cross-protection, including IFI44L, one of the strongest correlates, whose antiviral function remains poorly defined. These findings highlight how adenovirus infections spread and alter epithelial responses, demonstrate the ability of respiratory viruses to induce cross-protection, and identify ISGs—particularly IFI44L—as potential key mediators. This opens perspectives for future studies to dissect the antiviral role of IFI44L and evaluate its potential as a therapeutic target to enhance antiviral immunity in the respiratory tract.
ED Sociétés, Politique, Santé Publique
Regularization methods for high-dimensional data integration into mechanistic models : application for vaccine development.
by Auriane GABAUT (Bordeaux Population Health Research Center)
The defense will take place at h00 - Amphi PA Louis ISPED, Université de Bordeaux - Campus Carreire, 146 Rue Léo Saignat, 33000 Bordeaux
in front of the jury composed of
- Mélanie PRAGUE - Chargée de recherche - Université de Bordeaux - Directeur de these
- Emmanuelle COMETS - Chargée de recherche - INSERM Paris - Rapporteur
- Julien CHIQUET - Directeur de recherche - Université Paris-Saclay, INRAE, AgroParisTech - Rapporteur
- Pierre NEUVIAL - Directeur de recherche - CNRS, Institut de Mathématiques de Toulouse - Examinateur
- Sébastien BENZEKRY - Chargé de recherche - Inria Sophia-Antipolis , Inserm U1068, CNRS UMR7258 - Examinateur
- Cécile PROUST-LIMA - Directrice de recherche - Bordeaux Population Health Research Center, Inserm, Université de Bordeaux - CoDirecteur de these
- Hélène JACQMIN-GADDA - Directrice de recherche - Bordeaux Population Health Research Center, Inserm, Université de Bordeaux - Examinateur
The first part of the thesis focuses on the integration of transcriptomic markers measured at a fixed time point, used as explanatory variables for the inter-individual variability in the mechanistic model parameters. The proposed method is based on an iterative algorithm combining Lasso-type penalized regression and parameter estimation via the SAEM algorithm, in order to build a final model that is both parsimonious and interpretable. The second part addresses the integration of longitudinal transcriptomic data. It is based on the assumption that certain biomarkers — in particular gene expression profiles — reflect the dynamics of unobserved compartments in the mechanistic model, such as cellular populations. A coordinate descent algorithm is proposed, alternating between Lasso regularization of the biomarkers and population parameter estimation via SAEM, aiming to maximize a penalized joint likelihood. This approach allows simultaneous biomarker selection and parameter estimation. These two contributions have led to R packages available on CRAN or GitHub. The proposed methods are applied to vaccine trial data for Ebola, varicella, and SARS-CoV-2, to identify which transcriptomic markers best describe the establishment of the immune response following vaccination. The methodology can be more broadly applied to other complex biological processes, including diseases such as cancer, for which mechanistic models and high-dimensional data are available.