Embedded Files
Network dynamics during motor learning and re-learning with Parkinson’s Disease
Learning is a dynamic process driven by synaptic plasticity changes where neuronal activity adapts over time to improve performance or acquire new skills. Parkinson’s Disease (PD) is a neurodegenerative condition characterized by a loss of dopaminergic neurons in the basal ganglia which disrupts motor-related circuits in the primary motor cortex (M1) and compromises learned skills and motor control. In addition, Dopamine depletion affects the network's ability to efficiently adapt to new tasks, making re-learning challenging.
In this work, we develop analysis methods to understand the transformative process neuronal networks undergo during learning in healthy conditions and re-learning in PD. We base our research on experimental data (2-photon Calcium imaging data of behaving animals) and theoretical models for learning networks. Our analysis shows significant differences in the way the network reorganizes during learning with or without PD, concerning the way behavior is represented by the ensemble activity and the overall correlational configuration of the network.
Moving forward, our research focuses on developing novel methods to detect and characterize the specific sub-networks responsible for learning and re-learning under PD conditions. We will formulate a model that represents re-learning as a stochastic iterative process affected by PD and propose artificial models that mimic this process. Our goal is to provide a deeper understanding of the network dynamics involved in PD and to develop strategies that may restore learning capabilities in affected networks. This work, in collaboration with the Schiller lab, highlights the potential for bridging the gap between basic research and therapeutic advancements in PD
Data-driven models for remapping of large-scale neuronal recordings
Border cells, neurons that encode proximity to environmental boundaries, are assumed to play a key role in spatial navigation within the hippocampal formation. Previous studies have demonstrated their relative stability across environments, though their contribution to spatial encoding is not yet fully understood. We hypothesized that spatial encoding of population activity would be preserved across distinct environments through a transformation of the spatial coordinate system. To test this, we conducted large-scale chronic Neuropixel recordings in the Subiculum of mice as they navigated pairs of different rooms. Analysis of population rate maps suggests that while the overall function of border cells was consistent, individual cells exhibited shifts in their firing positions. Notably, spatial maps of the population were maintained across rooms, often through an affine transformation (i.e. translation, stretching, and rotation)—suggesting a flexible yet stable mechanism for encoding space. To further investigate and quantify these findings, we propose a novel architecture for data-driven detection of remapping at the population level. Our two-stage pipeline consists of a position decoder and a remapping decoder. Although highly non-linear, our architecture allows us to extract a data-driven remapping of position representation at the population level. We demonstrate the robustness of the proposed analysis in decoding positional information from Subiculum ensemble activity across room pairs in 22 arenas. We show that indeed, the learned remapping decoders suggest an affine transformation of population encoding across rooms. These findings reveal a novel insight into the adaptability of border cell function, showing that while individual cells may undergo remapping, the collective population can still maintain stable spatial representations across environments. This suggests a flexible, robust spatial encoding mechanism that allows for dynamic adjustments without compromising the overall spatial representation.
Cortex-wide correlation dynamics during sensory learning
The foundation for learning is the remarkable ability of cortical networks for plastic changes, where interactions between neurons and cortical regions change to improve neural representation and task performance. However, it remains unknown whether changes in functional connectivity are driven by specific types of cortical neurons, and if these are linked to corresponding changes in cortical activity levels.
We therefore used widefield calcium imaging to track the cortical dynamics of intratelencephalic (IT) and pyramidal tract (PT) neurons in transgenic mice learning an auditory discrimination task. While cortical activity in both groups increased during learning, particularly in parietal and frontal regions, these changes in activity were variable and not strongly correlated with task performance or learning progression. To study changes in the correlation structure, we computed correlation matrices between functionally-identified regions and extracted a low-dimensional representation based on Riemannian distances. We found a highly structured trajectory across time, indicating a gradual transformation of the correlational configuration throughout learning. Correlation structure was highly predictive of the learning stage and task performance, outperforming activity-based models, consistent across different trial periods, and both IT and PT neurons.
To further explore the dynamics of these connectivity changes we applied a Riemannian-Wavelet analysis, to identify cortical network components driving temporal changes in the correlation structure. We found that slow changes were predominantly driven by the parietal cortex, while fast changes were more variable, with PT neurons showing more consistent contributions from both parietal and frontal regions. In contrast, IT neurons showed greater spatial variability in their network interactions.
Overall, this study demonstrates that the evolution of cortical correlation structure, rather than just neuronal activity, provides critical insights into the neural mechanisms underlying learning. Our computational approach highlights the importance of network-level changes and identifies distinct dynamics in IT and PT neuron contributions to cortical plasticity.
The dynamics of learning - temporal evolution of SPD matrices
How does brain connectivity evolve through learning? This is the biological motivation for this project.
We explore the dynamics of correlations between neuronal activity of cells /brain regions and how they evolve to create new representations and improve performance during learning. Diverging from the commonly used approaches relying on statistics, we use a geometric point of view of correlations as points on a Riemannian manifold. We will develop a unique approach to model the temporal evolution of correlation matrices throughout the learning process. This interpretable approach would allow us to identify the intrinsic time scales of the data as well as the main components driving this process.
The contribution of this project will be twofold: a mathematical framework for modeling temporal dynamics of SPD matrices, and biological - promoting a deeper understanding of functional connectivity in the brain.
Google Sites
Report abuse