Periodic Reporting for period 2 - DYNAMIC_ENGRAM (Deciphering the enigma of memory persistence: how the brain stably stores information using dynamic networks and unstable neurons)
Okres sprawozdawczy: 2022-08-01 do 2024-01-31
The DYNAMIC_ENGRAM project deals with a fundamental question in neuroscience: how does the brain store and retrieve information over time? The accepted notion that these processes rely on the neuronal ensembles that were active during learning is now challenged by findings by our lab and others that reveal that different neurons and networks than those that were active during learning support persistent memory. Most notably, we found that the long-term persistence of spatial memory is correlated with the degree to which neuronal activity is spatially informative, but not with the stability of the coding carried by individual neurons. These discoveries expose a fundamental gap in our understanding and highlight the need to reveal how neural codes across brain circuits, including the hippocampus, entorhinal cortex, and prefrontal cortex, change over the lifetime of a memory.
In this project, we investigate the mechanisms that govern the reorganization of memory using innovative methods we recently developed for optical imaging, large-scale data analysis, and circuit manipulation. Key among them is our ability to simultaneously and longitudinally image in two related brain areas the activity of large neuronal populations in freely behaving mice. Using these new tools, we will elucidate the factors governing the circuit dynamics of memory representations (Aim 1); how such dynamics relate to the behavioral manifestation of memory (Aim 2); how hippocampal-cortical and cortical-cortical interactions change over weeks to support remote memory (Aim 3); and what mechanisms could underlie the transfer of learned information between neurons in a network (Aim 4).
Our working hypothesis is that long-term memory is supported by a dynamic (rather than a static) engram, in which information is retained at the population-level and individual neurons are (at least partially) interchangeable. Specifically, we aim to examine the possibility that the stability of the internal structure (manifold) of neuronal population activity can better explain the persistence of memory than the stability of the tuning of individual neurons. If confirmed, this notion may revise current theories in the field by showing how long-lived memories can be supported by a dynamic engram.
In this project, we investigate the mechanisms that govern the reorganization of memory using innovative methods we recently developed for optical imaging, large-scale data analysis, and circuit manipulation. Key among them is our ability to simultaneously and longitudinally image in two related brain areas the activity of large neuronal populations in freely behaving mice. Using these new tools, we will elucidate the factors governing the circuit dynamics of memory representations (Aim 1); how such dynamics relate to the behavioral manifestation of memory (Aim 2); how hippocampal-cortical and cortical-cortical interactions change over weeks to support remote memory (Aim 3); and what mechanisms could underlie the transfer of learned information between neurons in a network (Aim 4).
Our working hypothesis is that long-term memory is supported by a dynamic (rather than a static) engram, in which information is retained at the population-level and individual neurons are (at least partially) interchangeable. Specifically, we aim to examine the possibility that the stability of the internal structure (manifold) of neuronal population activity can better explain the persistence of memory than the stability of the tuning of individual neurons. If confirmed, this notion may revise current theories in the field by showing how long-lived memories can be supported by a dynamic engram.
Work on this project has been progressing in parallel on all four specific aims. We have published our findings thus far in four papers.
The first project directly addresses Aim 1 of the ERC project: to determine how time and experience interact to change spatial representations. In this project (Geva et al, Neuron 2023) we found that time and experience differentially affect distinct aspects of hippocampal representational drift. Specifically, the passage of time and the amount of experience are two factors that profoundly affect memory, but thus far, it has remained unclear to what extent these factors drive hippocampal representational drift. We longitudinally recorded large populations of hippocampal neurons in mice while they repeatedly explored two different familiar environments that they visited at different time intervals over weeks. We found that time and experience differentially affected distinct aspects of representational drift: the passage of time drove changes in neuronal activity rates, whereas experience drove changes in the cells’ spatial tuning. Changes in spatial tuning were context specific and largely independent of changes in activity rates. Thus, our results suggest that representational drift is a multi-faceted process governed by distinct neuronal mechanisms.
Additional work aimed to elucidate the mechanisms that govern neural code stability. We reasoned that neural network architecture and connectivity might play a crucial role in determining neural code stability. Thus, we compared between two hippocampal circuits that are known to have a distinct network architecture and are thought to play different roles in long-term memory. Specifically, we compared neural code quality and stability between the CA1 and its upstream circuit the CA3. In this project (Sheituch et al, Cell Reports 2023) we found that the organization of hippocampal CA3 into correlated cell assemblies supports a stable spatial code. Briefly, the CA3 is thought to stably store memories in assemblies of recurrently connected cells functioning as a collective, but the collective hippocampal coding properties that are unique to CA3 and how such properties facilitate the stability or precision of the neural code remained unclear. We performed large-scale Ca2+ imaging in the CA1 and CA3 of freely behaving mice that repeatedly explored the same, initially novel environments over weeks. We found that CA3 place cells have more precise and more stable tuning and show a higher statistical dependence with their peers compared with CA1 place cells, uncovering a cell assembly organization in CA3. Surprisingly, although tuning precision and long-term stability are correlated, cells with stronger peer dependence exhibit higher stability but not higher precision. These results expose the three-way relationship between tuning precision, long-term stability, and tuning peer dependence.
Consistent with the above results, we found (Deitch, Rubin and Ziv, Current Biology, 2021) that representational drift occurs not only in deep or associative brain areas such the hippocampus, but also in sensory areas such as the visual cortex. In this work, we analyzed large-scale optical and electrophysiological recordings from six visual cortical areas in behaving mice that were repeatedly presented with the same natural movies. Contrary to the prevailing notion, we found representational drift over timescales spanning minutes to days across multiple visual areas, cortical layers, and cell types. Surprisingly, neural-code stability did not reflect the hierarchy of information flow across areas. Although individual neurons showed time-dependent changes in their coding properties, the structure of the relationships between population activity patterns remained stable and stereotypic. These results are consistent with our working hypothesis (detailed above), suggesting that the observed population-level organization may underlie stable perception despite continuous changes in neuronal responses.
In addition to the above-mentioned discoveries, as part of our work to improve the analysis of the large amounts of data we acquire in our experiments, we have also advanced a methodology that is commonly used in systems neuroscience research. Neuroscientists interested in understanding the nature of the neural code often apply methods derived from the mathematical framework of information theory to quantify the statistical relationship between neuronal activity and a certain variable of interest. For studying the neural basis for spatial navigation, it is useful to estimate how much information hippocampal neurons carry about the position of an animal within a specific environment. However, the standard measures for estimating information content suffer from an upward bias when applied to small sample sizes, which may lead to misinterpretation of the data. This bias is more pronounced in data from calcium imaging–a widely used technique for recording neuronal activity–because the activity extracted from the measured calcium signal is sparse in time. In this project (Sheintuch, Rubin, and Ziv, Plos. Comp. Biol. 2022) we introduce new methods to correct the bias in the naïve estimation of information content from limited sample sizes and temporally sparse neuronal activity. We show that our bias-correction methods allow an accurate estimation of the information content carried by the activity obtained from calcium imaging data in both hippocampal and cortical neurons, and help uncover differences in the way information content changes during learning across neural circuits.
The first project directly addresses Aim 1 of the ERC project: to determine how time and experience interact to change spatial representations. In this project (Geva et al, Neuron 2023) we found that time and experience differentially affect distinct aspects of hippocampal representational drift. Specifically, the passage of time and the amount of experience are two factors that profoundly affect memory, but thus far, it has remained unclear to what extent these factors drive hippocampal representational drift. We longitudinally recorded large populations of hippocampal neurons in mice while they repeatedly explored two different familiar environments that they visited at different time intervals over weeks. We found that time and experience differentially affected distinct aspects of representational drift: the passage of time drove changes in neuronal activity rates, whereas experience drove changes in the cells’ spatial tuning. Changes in spatial tuning were context specific and largely independent of changes in activity rates. Thus, our results suggest that representational drift is a multi-faceted process governed by distinct neuronal mechanisms.
Additional work aimed to elucidate the mechanisms that govern neural code stability. We reasoned that neural network architecture and connectivity might play a crucial role in determining neural code stability. Thus, we compared between two hippocampal circuits that are known to have a distinct network architecture and are thought to play different roles in long-term memory. Specifically, we compared neural code quality and stability between the CA1 and its upstream circuit the CA3. In this project (Sheituch et al, Cell Reports 2023) we found that the organization of hippocampal CA3 into correlated cell assemblies supports a stable spatial code. Briefly, the CA3 is thought to stably store memories in assemblies of recurrently connected cells functioning as a collective, but the collective hippocampal coding properties that are unique to CA3 and how such properties facilitate the stability or precision of the neural code remained unclear. We performed large-scale Ca2+ imaging in the CA1 and CA3 of freely behaving mice that repeatedly explored the same, initially novel environments over weeks. We found that CA3 place cells have more precise and more stable tuning and show a higher statistical dependence with their peers compared with CA1 place cells, uncovering a cell assembly organization in CA3. Surprisingly, although tuning precision and long-term stability are correlated, cells with stronger peer dependence exhibit higher stability but not higher precision. These results expose the three-way relationship between tuning precision, long-term stability, and tuning peer dependence.
Consistent with the above results, we found (Deitch, Rubin and Ziv, Current Biology, 2021) that representational drift occurs not only in deep or associative brain areas such the hippocampus, but also in sensory areas such as the visual cortex. In this work, we analyzed large-scale optical and electrophysiological recordings from six visual cortical areas in behaving mice that were repeatedly presented with the same natural movies. Contrary to the prevailing notion, we found representational drift over timescales spanning minutes to days across multiple visual areas, cortical layers, and cell types. Surprisingly, neural-code stability did not reflect the hierarchy of information flow across areas. Although individual neurons showed time-dependent changes in their coding properties, the structure of the relationships between population activity patterns remained stable and stereotypic. These results are consistent with our working hypothesis (detailed above), suggesting that the observed population-level organization may underlie stable perception despite continuous changes in neuronal responses.
In addition to the above-mentioned discoveries, as part of our work to improve the analysis of the large amounts of data we acquire in our experiments, we have also advanced a methodology that is commonly used in systems neuroscience research. Neuroscientists interested in understanding the nature of the neural code often apply methods derived from the mathematical framework of information theory to quantify the statistical relationship between neuronal activity and a certain variable of interest. For studying the neural basis for spatial navigation, it is useful to estimate how much information hippocampal neurons carry about the position of an animal within a specific environment. However, the standard measures for estimating information content suffer from an upward bias when applied to small sample sizes, which may lead to misinterpretation of the data. This bias is more pronounced in data from calcium imaging–a widely used technique for recording neuronal activity–because the activity extracted from the measured calcium signal is sparse in time. In this project (Sheintuch, Rubin, and Ziv, Plos. Comp. Biol. 2022) we introduce new methods to correct the bias in the naïve estimation of information content from limited sample sizes and temporally sparse neuronal activity. We show that our bias-correction methods allow an accurate estimation of the information content carried by the activity obtained from calcium imaging data in both hippocampal and cortical neurons, and help uncover differences in the way information content changes during learning across neural circuits.
We are proceeding according to the plan outlined in the project’s proposal. Specifically, we have further refined the imaging preparation. We are currently performing experiments that will serve to test the hypothesis that the functional connectivity between hippocampus and cortical areas increases with learning, and is correlated with memory performance at all times (as predicted by the multiple-trace and memory-indexing theories); and the alternative hypothesis, that functional connectivity between hippocampus and cortical areas is higher during recent recall than during remote recall (as predicted by the standard consolidation model).
Our experiments will also test the hypotheses that through their association with reactivated memory representations, hippocampal neurons could acquire tuning to an aspect of an experienced event without actually participating in the encoding of such an event during memory acquisition.
Our experiments will also test the hypotheses that through their association with reactivated memory representations, hippocampal neurons could acquire tuning to an aspect of an experienced event without actually participating in the encoding of such an event during memory acquisition.