Interference-Based Forgetting in Spatial Navigation: Latent Behavioral Signatures and the Impact of Synaptic Plasticity Disruption
[eng] Memory is a core adaptive function that enables animals to anticipate outcomes and guide behavior beyond immediate sensory cues. In this thesis, we use the 8-port maze (Morales et al., 2020) as a paradigm to disentangle learning from memory, to track the coexistence of multiple temporal traces...
| Autor: | |
|---|---|
| Tipo de recurso: | tesis doctoral |
| Estado: | Versión publicada |
| Fecha de publicación: | 2025 |
| País: | España |
| Institución: | Universidad de Barcelona |
| Repositorio: | Dipòsit Digital de la UB |
| OAI Identifier: | oai:diposit.ub.edu:2445/224673 |
| Acceso en línea: | https://hdl.handle.net/2445/224673 http://hdl.handle.net/10803/696025 |
| Access Level: | acceso abierto |
| Palabra clave: | Neurociències Neurociència computacional Memòria Aprenentatge automàtic Autoanticossos Neurosciences Computational neuroscience Memory Machine learning Autoantibodies |
| Sumario: | [eng] Memory is a core adaptive function that enables animals to anticipate outcomes and guide behavior beyond immediate sensory cues. In this thesis, we use the 8-port maze (Morales et al., 2020) as a paradigm to disentangle learning from memory, to track the coexistence of multiple temporal traces, and to dissect the mechanisms of memory interference and strategy dynamics in recall. In Chapter 1, we demonstrate that learning performance during training does not predict recall strength. Using the Memory Index (MI), we establish a robust and training-independent measure of memory, showing that animals recall memories from both 2h and 24h delays within the same session, thus providing behavioral evidence that multiple temporal traces coexist rather than overwrite one another. In Chapter 2, we uncover a spatial distance-dependent pattern of interference: memories from nearby rewarded locations reinforce one another, whereas distant ones interfere and weaken recall. A generative model partially reproduced this effect, suggesting that spatial relationships between memories contribute to, but do not fully account for, their stability. In Chapter 3, we apply a GLM-HMM to identify the latent strategies underlying recall. Four distinct strategies emerge—goal-directed recall (2h), outdated recall (24h), memory-avoidance, and lapses/disengagement. These strategies are expressed dynamically within and across sessions, explaining variability in behavior beyond average performance. Importantly, the spatial patterns of latent strategies show that interference is not a single, uniform effect, but becomes evident through the use of particular states, especially those guided by memory. In Chapter 4, we show that anti-NMDAR antibody infusion produces a reorganization of strategy dynamics: recall behavior becomes more dominated by 2h strategies at the expense of flexibility, and the spatial interference pattern of memory-based strategies is inverted. In Chapter 5, we find that anti-LGI1 antibodies also disrupt recall dynamics but in a qualitatively different manner: increases in 2h recall are more modest, while 24h recall and avoidance strategies persist in altered proportions, and interference spatial modulation shows non-monotonic patterns. These results reveal distinct disruptions of memory organization across autoimmune encephalitis models. Taken together, this thesis supports—based on behavioral, computational, and experimental evidence—that memory recall is not a fixed process, but can emerge from the flexible combination of multiple latent strategies. The 8-port maze, combined with latent-state modeling, offers a useful framework to bridge synaptic dysfunction and behavioral memory organization, while also opening the door to correlating behavior with neuronal imaging data. |
|---|