Periodic Reporting for period 1 - BrownianReactivation (Neural stochasticity and criticality in memory replay)
Berichtszeitraum: 2019-11-01 bis 2021-10-31
Our main goal is to elucidate the role of the dynamical state of cortical neural networks in facilitating memory consolidation. Such process is central to integrating novel information into existing knowledge and to the ability to generalize from single experiences. It is now accepted that in the brain the interplay between hippocampus and neo-cortex is central to system consolidation. The two structures are generally assigned with complementary roles, with the hippocampus in charge of short-term and episodic memories and the cortex of long-term memories with more pronounced semantic nature. Nevertheless, very little is known about the details and the modalities of this interaction and about the principles governing such reorganization of information.
In this project we aim at investigating whether activity propagation in cortical networks has properties which are particularly well suited to create long-range interactions across a large number of networks with heterogeneous functions and specializations. Such property would be functional to integrate complex, detail-rich novel experiences. In particular we ask whether such large-scale recruitment of specialized cortical networks happens in coincidence with peaks in the activity of the hippocampus, a region known to be central in the formation of novel memory and in supporting future planning.
We thus investigated this interaction between the content of hippocampal activations both during wakefulness and sleep and that of cortical avalanches, using a combination of data analysis and mathematical modelling.
While furthering our understanding of memory processes can be of great importance in addressing memory and learning impairments and neurodegenerative disorders in humans, this study also holds close ties with the fast-expanding field of Artificial Intelligence and Machine Learning. In fact, the functioning of the brain has been the main source of inspiration for the development of modern information processing systems. Modern challenges to Artificial Intelligence, mostly cantered on its ability to generalize and to extract regularities from the world without any form of supervision, are closely related to the open issues we are addressing here in the context of a biological information processing system as the central nervous system. By unveiling evolution-driven solutions adopted by the brain, our research has thus also the potential to provide novel directions in the field of artificial neural networks.
In this project we aim at investigating whether activity propagation in cortical networks has properties which are particularly well suited to create long-range interactions across a large number of networks with heterogeneous functions and specializations. Such property would be functional to integrate complex, detail-rich novel experiences. In particular we ask whether such large-scale recruitment of specialized cortical networks happens in coincidence with peaks in the activity of the hippocampus, a region known to be central in the formation of novel memory and in supporting future planning.
We thus investigated this interaction between the content of hippocampal activations both during wakefulness and sleep and that of cortical avalanches, using a combination of data analysis and mathematical modelling.
While furthering our understanding of memory processes can be of great importance in addressing memory and learning impairments and neurodegenerative disorders in humans, this study also holds close ties with the fast-expanding field of Artificial Intelligence and Machine Learning. In fact, the functioning of the brain has been the main source of inspiration for the development of modern information processing systems. Modern challenges to Artificial Intelligence, mostly cantered on its ability to generalize and to extract regularities from the world without any form of supervision, are closely related to the open issues we are addressing here in the context of a biological information processing system as the central nervous system. By unveiling evolution-driven solutions adopted by the brain, our research has thus also the potential to provide novel directions in the field of artificial neural networks.
First of all, we studied how hippocampal activation patterns can temporally organize to dynamically encode information either about current sensory information, or past experience. By studying the activity of so called ‘place cells’ while a mouse was freely exploring an environment, we found that these cells could modulate their activity in response to external input and local network processes, to form sequential patterns suited to either memory encoding or retrieval (Figure 1). Moreover, different groups of cells would be preferentially dedicated to one of the two functions, hinting at the existence of parallel information streams in the hippocampus, balancing the creation of novel memory traces and the maintenance of already existing ones. Current analysis of subsequent sleep activity aims at elucidate the role of these neurons in the consolidation of memory.
Looking at larger context of hippocampal-cortical dialogue, we took advantage of state-of-the-art imaging data of the entire cortical surface and applied advanced statistical methods to develop a model of cortical spontaneous activity propagation during sleep. We delineated the presence of multiple integrated functional networks, hierarchically organized and spatially segregated (Figure 2). Such networks largely overlapped with known subdivisions based on cortical connectivity and in particular appeared to indicate the existence of a set of networks (akin to the Default Mode Network found in humans) preferentially communicating with the hippocampus and actually controlling its activation during periods of rest. Interestingly these networks displayed many of the signatures of a critical state, even more strongly the closer was their interaction with the hippocampus, a finding that can have important implications for the processing of mnemonic information.
These results are now in the process of being published in high-profile peer reviewed journals and are the subject of multiple presentations at conferences and scientific meetings.
To better understand these implications, we are currently complementing these analyses on experimental data with a set of modeling studies aimed both at studying the importance of the temporal structure of hippocampal place cell activity, and at investigating the dynamics of learning in a neural network operating at a critical transition point.
Looking at larger context of hippocampal-cortical dialogue, we took advantage of state-of-the-art imaging data of the entire cortical surface and applied advanced statistical methods to develop a model of cortical spontaneous activity propagation during sleep. We delineated the presence of multiple integrated functional networks, hierarchically organized and spatially segregated (Figure 2). Such networks largely overlapped with known subdivisions based on cortical connectivity and in particular appeared to indicate the existence of a set of networks (akin to the Default Mode Network found in humans) preferentially communicating with the hippocampus and actually controlling its activation during periods of rest. Interestingly these networks displayed many of the signatures of a critical state, even more strongly the closer was their interaction with the hippocampus, a finding that can have important implications for the processing of mnemonic information.
These results are now in the process of being published in high-profile peer reviewed journals and are the subject of multiple presentations at conferences and scientific meetings.
To better understand these implications, we are currently complementing these analyses on experimental data with a set of modeling studies aimed both at studying the importance of the temporal structure of hippocampal place cell activity, and at investigating the dynamics of learning in a neural network operating at a critical transition point.
The results of the project so far are already calling for a profound re-evaluation of the foundations of brain-wide information transmission during sleep periods. We have unveiled a major association between cortical transients originating from a specific network subset and activation bouts in the hippocampus. As we also show cortical activity precedes hippocampal one, our results challenge the established idea that during sleep the information almost uniquely flow from the hippocampus to the cortex, mostly in association with so called Sharp Wave Ripple events in the hippocampus. Moreover, our analysis also indicate that cortex operates very close to a dynamical regime which optimizes the communication between far-apart regions and between networks specialized in processing different types of information, and that this state is specific to certain periods of sleep but disappears when the animal is awake. We thus hypothesize that it developed to maximize the interaction between networks during offline periods while these networks operate independently during active experience.
These results are compounded by an in-depth analysis of hippocampal temporal coding schemes which identified a novel subdivision in cell populations dynamically encoding either memory- or sensory-driven information. This project will further explore the consequence of such subdivision on sleep-related activations. Furthermore we will integrate our findings in a set of mathematical models of the hippocampal circuit and of its interactions with other parts of the cortex. These modelling studies will be allow us to demonstrate the implications of the neural dynamics we have described so far on the functioning of a brain-wide memory processing and learning system. The derived insight can be easily transferred to non-biological learning systems, which might benefit from being based on similar organizational as the biological brain.
These results are compounded by an in-depth analysis of hippocampal temporal coding schemes which identified a novel subdivision in cell populations dynamically encoding either memory- or sensory-driven information. This project will further explore the consequence of such subdivision on sleep-related activations. Furthermore we will integrate our findings in a set of mathematical models of the hippocampal circuit and of its interactions with other parts of the cortex. These modelling studies will be allow us to demonstrate the implications of the neural dynamics we have described so far on the functioning of a brain-wide memory processing and learning system. The derived insight can be easily transferred to non-biological learning systems, which might benefit from being based on similar organizational as the biological brain.