Final Report Summary - DEMORY (Dissecting the Role of Dendrites in Memory)
A grand challenge in neuroscience is to understand the mechanisms underlying memory functions. While the key role of dendrites in such functions is recognized, a unifying theory of how dendrites are utilized to achieve memory encoding, storage and/or retrieval is missing. The dEMORY project aimed at (a) developing computational models of single cells, micro-circuits and large neuronal networks of three brain areas: the hippocampus, the prefrontal cortex and the amygdala, (b) using the models to characterize the functional role of dendrites in memory processes in these regions and (c) deducing theoretical models that incorporate the most critical dendritic attributes, in an attempt to unify dendritic function in memory across brain areas. The project was highly successful in all of its aims:
Role of dendrites in hippocampal functions.
We examined and mapped the role of dendrites in hippocampal processing at both the circuit and network levels. Our DG circuit models predicted how granule cell dendrites contribute to pattern separation (Chavlis and Poirazi, 2017; Chavlis et al., 2017) and revealed how Mossy Cells aid context discrimination (Danielson et al., 2017). Together with experimental collaborators, we developed CA1 circuit models to explain how VIP interneurons, which control dendritic excitability, contribute to spatial learning (Turi*, Li*, Chavlis* et al, under revision in Neuron) and how epilepsy impacts such spatial learning (Shuman et al., BioRxiv 2018 & under revision). Extending to the region levels, we developed network models of the hippocampus and showed that dendritically targeted inhibition is also critical for shaping the activity latencies observed in CA1 pyramidal neurons of behaving animals (Cutsuridis and Poirazi, 2015).
Role of dendrites in associative memory formation (amygdala/hippocampus)
We developed a large scale network model of simplified pyramidal neurons with nonlinear dendrites, incorporating a wide range of synaptic and intrinsic plasticity rules (Kastellakis et al., 2016). The model predicts that dendritic non-linearities of both pyramidal and interneurons (Tzilivaki, Kastellakis and Poirazi, under revision) and cooperative synaptic plasticity underlie the formation of synaptic clusters within dendrites, which in turn form the backbone for information binding and storage capacity (Kastellakis et al., 2016; Frank et al., 2017). These findings support our novel hypothesis whereby clusters of synapses within non-linear dendrites serve as the smallest processing unit underlying memory formation (Kastellakis et al., 2015).
Role of dendrites in prefrontal cortex functions
We probed dendritic integration and memory across levels and neuron types. Specifically, we developed a detailed biophysical model of Fast Spiking Basket cells and showed for the first time that the dendrites of these interneurons integrate their inputs in a highly non-linear manner (Tzilivaki and Poirazi, bioRxiv 2018). At the circuit level, we developed biophysically detailed microcircuit models of PFC neurons and showed that dendritic nonlinearities (e.g. NMDA and other ionic conductances (Papoutsi et al., 2013, 2014, 2017) as well as different types of inhibition (Konstantoudaki et al., 2014) are critical for the induction, stability and specificity of persistent activity, the cellular correlate of working memory. To facilitate the analysis of simulated data, we also developed novel theoretical methods for assessing specificity in these networks (Petrantonakis and Poirazi, 2015). Ongoing work (Stefanou et al, in preparation, 2018) reveals how different connectivity schemes (random or patterned) may influence signal encoding in large-scale networks of morphologically and biophysically realistic PFC microcircuits.
The role of dendritic morphology in shaping the activity profiles of L5 PFC pyramidal neurons has also been investigated with computational models (Psarrou et al., 2014; Papoutsi et al., 2017). A novel tool that allows the modification of any neuronal morphology has been developed (Bozelos et al., 2016) so as to explore the causal link between dendritic morphology and neuronal output at both the single cell and network levels. Inspired by the above findings, we further sought to validate the critical role of dendrites in the neocortex. Together with collaborators, we investigated the role of apical and basal trees in orientation tuning in L2/3 neurons of the visual cortex in anesthetized mice (Park et al, under revision).
Finally, with respect to the deduction of mathematical abstractions, we generated the first reduced models of fast spiking basket cells with non-linear dendrites (Tzilivaki, Kastellakis and Poirazi, under revision). At the brain area level, we explore a radically new hypothesis according to which hippocampal subregions map onto different steps of Compressed Sensing, a recent advancement in the signal-processing field (Petrantonakis and Poirazi, 2014). Towards proving this hypothesis, we showed that incorporation of the circuitry features of DG can improve the performance of a state of the art Compressed Sensing algorithm (Petrantonakis and Poirazi, 2015b).
Overall, both computational models and findings of the dEMORY project are expected to have a major impact on our understanding of how dendrites contribute to memory formation in the brain. They reveal common features of processing underlying different memory functions across brain areas and enable the extraction of common rules and theoretical models that can be used in fields such as artificial intelligence and machine learning. Our work also provides new, publicly available tools and methodologies for studying and manipulating dendritic features at different levels and functionalities, thus facilitating further research in this field.
Role of dendrites in hippocampal functions.
We examined and mapped the role of dendrites in hippocampal processing at both the circuit and network levels. Our DG circuit models predicted how granule cell dendrites contribute to pattern separation (Chavlis and Poirazi, 2017; Chavlis et al., 2017) and revealed how Mossy Cells aid context discrimination (Danielson et al., 2017). Together with experimental collaborators, we developed CA1 circuit models to explain how VIP interneurons, which control dendritic excitability, contribute to spatial learning (Turi*, Li*, Chavlis* et al, under revision in Neuron) and how epilepsy impacts such spatial learning (Shuman et al., BioRxiv 2018 & under revision). Extending to the region levels, we developed network models of the hippocampus and showed that dendritically targeted inhibition is also critical for shaping the activity latencies observed in CA1 pyramidal neurons of behaving animals (Cutsuridis and Poirazi, 2015).
Role of dendrites in associative memory formation (amygdala/hippocampus)
We developed a large scale network model of simplified pyramidal neurons with nonlinear dendrites, incorporating a wide range of synaptic and intrinsic plasticity rules (Kastellakis et al., 2016). The model predicts that dendritic non-linearities of both pyramidal and interneurons (Tzilivaki, Kastellakis and Poirazi, under revision) and cooperative synaptic plasticity underlie the formation of synaptic clusters within dendrites, which in turn form the backbone for information binding and storage capacity (Kastellakis et al., 2016; Frank et al., 2017). These findings support our novel hypothesis whereby clusters of synapses within non-linear dendrites serve as the smallest processing unit underlying memory formation (Kastellakis et al., 2015).
Role of dendrites in prefrontal cortex functions
We probed dendritic integration and memory across levels and neuron types. Specifically, we developed a detailed biophysical model of Fast Spiking Basket cells and showed for the first time that the dendrites of these interneurons integrate their inputs in a highly non-linear manner (Tzilivaki and Poirazi, bioRxiv 2018). At the circuit level, we developed biophysically detailed microcircuit models of PFC neurons and showed that dendritic nonlinearities (e.g. NMDA and other ionic conductances (Papoutsi et al., 2013, 2014, 2017) as well as different types of inhibition (Konstantoudaki et al., 2014) are critical for the induction, stability and specificity of persistent activity, the cellular correlate of working memory. To facilitate the analysis of simulated data, we also developed novel theoretical methods for assessing specificity in these networks (Petrantonakis and Poirazi, 2015). Ongoing work (Stefanou et al, in preparation, 2018) reveals how different connectivity schemes (random or patterned) may influence signal encoding in large-scale networks of morphologically and biophysically realistic PFC microcircuits.
The role of dendritic morphology in shaping the activity profiles of L5 PFC pyramidal neurons has also been investigated with computational models (Psarrou et al., 2014; Papoutsi et al., 2017). A novel tool that allows the modification of any neuronal morphology has been developed (Bozelos et al., 2016) so as to explore the causal link between dendritic morphology and neuronal output at both the single cell and network levels. Inspired by the above findings, we further sought to validate the critical role of dendrites in the neocortex. Together with collaborators, we investigated the role of apical and basal trees in orientation tuning in L2/3 neurons of the visual cortex in anesthetized mice (Park et al, under revision).
Finally, with respect to the deduction of mathematical abstractions, we generated the first reduced models of fast spiking basket cells with non-linear dendrites (Tzilivaki, Kastellakis and Poirazi, under revision). At the brain area level, we explore a radically new hypothesis according to which hippocampal subregions map onto different steps of Compressed Sensing, a recent advancement in the signal-processing field (Petrantonakis and Poirazi, 2014). Towards proving this hypothesis, we showed that incorporation of the circuitry features of DG can improve the performance of a state of the art Compressed Sensing algorithm (Petrantonakis and Poirazi, 2015b).
Overall, both computational models and findings of the dEMORY project are expected to have a major impact on our understanding of how dendrites contribute to memory formation in the brain. They reveal common features of processing underlying different memory functions across brain areas and enable the extraction of common rules and theoretical models that can be used in fields such as artificial intelligence and machine learning. Our work also provides new, publicly available tools and methodologies for studying and manipulating dendritic features at different levels and functionalities, thus facilitating further research in this field.