ED Sciences de la Vie et de la Santé
Is Alzheimer's disease unique to the human species?
by Clara TOUSSAINT (Immunologie Conceptuelle, Expérimentale et Translationnelle)
The defense will take place at 14h00 - Amphithéâtre Université de Bordeaux, Bâtiment BBS, rue Docteur Hoffmann, 33000 Bordeaux
in front of the jury composed of
- Maël LEMOINE - Professeur des universités - Université de Bordeaux - Directeur de these
- Marie-Claude POTIER - Directrice de recherche - Sorbonne Université - Rapporteur
- Nicolas VILLAIN - Maître de conférences - praticien hospitalier - Sorbonne université, Paris 6 - Examinateur
- Lara KEUCK - Professeure des universités - Bielefeld University - Rapporteur
- Nora ABROUS - Directrice de recherche - Université de Bordeaux - Examinateur
- Vincent PLANCHE - Professeur des universités - praticien hospitalier - Université de Bordeaux - CoDirecteur de these
Alzheimer's disease (AD) is a major public health problem in our ageing Western societies. More than a century after it was first described, there is still no consensus on the nosology of AD. A biological perspective suggests that the mere presence of extracellular amyloid protein deposits and intracellular tau protein aggregates is sufficient to define the condition (Jack et al, 2018), these abnormalities being regarded as ‘causal' according to the amyloid cascade hypothesis. For other authors, biological findings alone are insufficient and must be accompanied by clinical signs suggestive of a progression towards dementia (Dubois et al, 2021), particularly as biological abnormalities may precede symptoms by 10 to 20 years, and many healthy older people with amyloid deposits will never develop the disease. This nosological issue causes the estimated global prevalence to vary from 32 million (dementia-based definition) to 315 million people (biological definition) (Gustavsson et al, 2023), with major implications for future diagnostic and therapeutic strategies (plasma biomarkers, anti-amyloid immunotherapies). This issue is reflected in the representation of AD within the animal kingdom. Certain characteristic lesions have been reported in rodents (Inestrosa et al, 2015), cetaceans (Vacher et al, 2022) and aged primates (Heuer et al, 2012), but these species do not develop all the characteristics of human tauopathy, and the cognitive impact of these lesions remains uncertain. The very concept of dementia, based on the loss of independence in everyday activities, remains an anthropocentric definition with no equivalent in the animal kingdom. Alzheimer's disease would therefore exist in animals according to a minimalist biological definition centred on amyloid (the AD continuum), but not according to a clinico-biological definition. The first part of this thesis addresses this question through a systematic review of the literature (38 articles, 543 non-human primate brains, four species). Using logistic regressions, this study shows that amyloid plaques follow isometric dynamics, proportional to the lifespan of each species, whilst NFTs follow chronometric dynamics, appearing at a fixed time point (30 to 50 years) regardless of life expectancy. This dissociation challenges the hypothesis of the amyloid cascade as a universal mechanism and suggests that full-blown AD only emerges when a species' lifespan allows these two independent processes to overlap. The second line of research focuses on monkeys that have received injections of pathological tau proteins (with or without co-injection of oligomeric Aβ). We hypothesise that a differential analysis of human and primate proteomes in response to these proteins will help to identify mechanisms of vulnerability specific to humans, owing to the phylogenetic proximity between macaques and humans. The results will be confirmed by Western blot or immunohistochemistry. The identification of resistance mechanisms in macaques, or vulnerability mechanisms in humans, could open up new therapeutic avenues. The third line of inquiry is based on the observation that longevity, whilst a necessary condition for the emergence of a fully-fledged AD, is not sufficient: those affected must also survive long enough for the disease to become apparent at a population level, which in humans depends on a social trait: care. We are developing a mathematical model of an age-structured population, simulated on an individual-centred basis, which combines demographic dynamics with the social allocation of care, to test whether a social structure based on care – independently of longevity – can explain the human specificity of the prevalence of AD.
ED Sciences Physiques et de l'Ingénieur
Neural network image registration for hybrid navigation
by Simon BERTRAND (Laboratoire de l'Intégration du Matériau au Système)
The defense will take place at 9h00 - Amphi JP Dom Laboratoire IMS, 351 Cours de la Libération, A31, 33405 Talence
in front of the jury composed of
- Lionel BOMBRUN - Professeur - Bordeaux Sciences Agro - Directeur de these
- Nelly PUSTELNIK - Directrice de recherche - ENS de Lyon - Rapporteur
- Abdourrahmane ATTO - Professeur - Polytech Annecy-Chambéry - Rapporteur
- Mathieu FAUVEL - Directeur de recherche - INRAE - Examinateur
Autonomous navigation systems rely heavily on inertial sensors to estimate a vehicle's position and orientation over time. A key limitation of these sensors is the gradual accumulation of measurement errors, which leads to a drift in the estimated trajectory. In practice, this drift is typically corrected using external positioning systems such as GPS. When such systems become unavailable, degraded, or unreliable, there is no straightforward way to reinitialize the inertial drift using inertial measurements alone. Within this context, this thesis investigates an image-based registration approach, where an onboard SAR image is aligned with a georeferenced optical image in order to estimate the geometric transformation between the two observations and recover an accurate position estimate. SAR-optical registration is nevertheless a challenging problem. Although both modalities describe the same physical scene, they are governed by fundamentally different sensing mechanisms: optical imagery depends on visible and near-infrared reflectance, whereas SAR measures radar backscatter, which is influenced by scene geometry, surface roughness, speckle noise, defocusing effects, and other modality-specific phenomena. These differences are further amplified by variations in seasonality, acquisition conditions, and geographic regions. In this context, this thesis proposes several contributions aimed at improving the accuracy, robustness, and reliability of dense multimodal registration methods for safety-critical navigation applications. The first contribution consists of the design of the MEOW/Europe dataset, a multi-year and multi-season collection of paired optical (Sentinel-2) and radar (Sentinel-1) images covering the European continent. This dataset is enriched with geographic, temporal, and environmental metadata and provides a benchmark for training and evaluating multimodal registration methods. It enables the systematic study of deep learning architectures dedicated to optical-SAR registration. A series of architectural design choices are further explored through model hybridization, including variations in network structures, similarity measures, and loss functions. The second contribution introduces RoADNet, a rigid registration model capable of jointly estimating translation and rotation between image pairs. The method is based on a structured cross-correlation mechanism that builds a dense similarity volume in the roto-translation space (x,y,theta). This formulation improves robustness to angular misalignments while introducing a trade-off between angular resolution, estimation accuracy, and computational cost. Finally, this thesis addresses the problem of uncertainty quantification in registration models. Dense similarity maps are transformed into sets of plausible solutions using conformal prediction methods, allowing explicit control of the risk level. This framework provides an interpretable measure of confidence: prediction sets remain compact when the estimation is reliable and expand in the presence of ambiguity or degraded observations. This opens the way to integrating confidence-aware localization into navigation systems. Extensions are also investigated, in particular through knowledge distillation approaches aimed at reducing model complexity under embedded constraints.
Diffusion of one-dimensional carbon nanotubes in a complex soft matter environment
by Rohit MANGALWEDHEKAR (Laboratoire Photonique, Numérique & Nanosciences)
The defense will take place at 14h00 - Amphitheatre Institut d'Optique d'Aquitaine Rue François Mitterrand CS30006 33400 Talence Cedex France
in front of the jury composed of
- Laurent COGNET - Directeur de recherche - Université de Bordeaux - Directeur de these
- Jean-Pierre DELVILLE - Directeur de recherche - Laboratoire Ondes et Matière d'aquitaine (LOMA) - Examinateur
- Jean-Baptiste MASSON - Directeur de recherche - Institut Pasteur - Rapporteur
- David HOLCMAN - Directeur de recherche - Institute of Biology of ENS (IBENS) - Rapporteur
- Ignacio IZEDDIN - Maître de conférences - ESPCI - Examinateur
- Stephane BANCELIN - Chargé de recherche - Université de Bordeaux - CoDirecteur de these
The transport of nanoscale objects in complex environments underpins a wide range of processes, from biomolecular activity in the extracellular spaces of tissues to fluid transport through filters, catalysts, and metal-organic frameworks. Understanding how particle geometry interacts with local architecture to influence diffusion and escape dynamics remains a fundamental challenge in complex biological and soft matter systems. While the anomalous diffusion of spherical probes in heterogeneous environments is well established, far less is known about how anisotropic, 1D molecules explore confined microenvironments, or how their length influences residence times and escape through narrow passages. In my thesis, I investigated the diffusion of anisotropic carbon nanotubes in a complex soft matter system. Nanotubes functionalized with color centers (CCNTs), which enhance their infrared luminescence, were used. As a biophysical model system, densely packed emulsion droplets were prepared: the aqueous interstitial regions between the droplets reproduce complex soft matter environments such as the brain extracellular spaces (ECS). Using a near-infrared single-molecule microscopy setup operating over an extended depth of field and in three dimensions, I then imaged individual CCNTs of two distinct size scales, differing by an order of magnitude, diffusing in aqueous medium within the biophysical model. Having observed subdiffusive and antipersistent motion for nanotubes of both size scales, we extended our analysis to understand the nature of diffusion in local geometries and to better characterize the scaling of diffusion properties with particle length. To connect local dynamics with global diffusion, escape from confined chambers was investigated. Surprisingly, despite a tenfold increase in particle length, long CCNTs require only about 1.4x longer to escape than short CCNTs. Moreover, this weak dependence contrasts sharply with simple theoretical predictions, which suggest that escape times should scale approximately linearly with particle length. To understand this discrepancy, we implemented Single Particle Orientation Tracking microscopy to directly measure the azimuthal and polar orientations of the particle. These measurements reveal that long CCNTs become rotationally constrained and aligned within narrow channels, reducing their orientational freedom while facilitating their transport through the network. Together, these results identify geometric frustration as a fundamental mechanism underlying anomalous transport in continuous structured environments. Even in the absence of binding interactions or discrete trapping sites, geometry alone is sufficient to generate subdiffusive and antipersistent dynamics through repeated wall-mediated interactions and intermittent confinement. Furthermore, particle anisotropy determines how this geometric frustration is resolved, enabling confinement-induced alignment to facilitate rather than hinder transport. These findings establish general physical principles governing the diffusion of anisotropic objects in complex media and provide insight into transport processes in biological, porous, and soft matter systems. The principles identified in this thesis provide a foundation for future investigations of transport in increasingly realistic complex environments. Beyond rigid carbon nanotubes, future studies may employ flexible and deformable nanoscale probes to determine how mechanical properties such as bending and conformational fluctuations influence diffusion under confinement. In parallel, more sophisticated model systems incorporating flow gradients, biomolecular interactions, and gel-like matrices could bridge the gap between simplified soft matter architectures and biological extracellular spaces. Such a bottom-up approach offers a promising route toward understanding the combined roles of particle geometry, mechanics, and environmental complexity in governing nanoscale transport.