Activity Of Neural Network In The Hippocampal CA1 Region Encoding Startling Fear Memory

Learning and memory of startle events are important brain functions for protecting animals and humans from dangers and threats, which are essential for their survival. Much recent evidence suggested that the hippocampus might play an importance role in learning and memory of fear conditioning. Lesion studies suggested that both contextual and 'long-duration' trace conditioning depended on the hippocampus. Therefore, the question is how the hippocampal neural network can encode, process, and integrate the external information during fear learning and memory. Due to technical limitations, little has been studied about the underlying mechanisms of fear learning and memory at the neural network level.The questions we concern are as follows: 1. how the hippocampal neural network encodes discrete startling episodes; 2. how the hippocampal neural network represents associative startling episodes, and how the representations can be integrated into existing neural network for further consolidation; 3. what the fundamental molecular substrate is for the modification of neural network during fear learning and memory; 4. whether we have potential applications from our observations. Our research is to study the activity of neural network when mice undergoing startling fear episodes by using in vivo ensemble recording techniques for unveiling the underlying encoding mechanisms of fear learning and memory.1. Activity of neural cliques in the hippocampus encoding intensity of startle episodesBy using our self-constructed light microdrive with up to 128-channel electrodes and headstages, we simultaneously recorded on average 254 individual units from the hippocampal CA1 regions in the seven mice, while subjecting them to different startles, such as metal sound, airpuff, shake, and drop. Our result showed that all startle-responsive neurons can be divided into two groups according to their responses to the intensity of stimuli: intensity-sensitive and -insensitive neurons. Intensity-sensitive neurons changed their firing rates in response to different intensity levels of stimuli; whereas, insensitive ones did not. These two types of neurons had different contributions to encoding the startle information: intensity-insensitive neurons played the major role in characterizing the basic features of startle episodes by answering the question what the startle was; intensity-sensitive neurons depicted detailed information for a certain startle episode by addressing the question how strong the startle was. Both types of neurons were organized in a categorical and hierarchical manner for encoding features of startle episodes and discriminating one startle from others; at the same time, their behaviors also followed the frequency code theory when the firing rates were modulated for encoding intensity information of a certain type of startle stimuli.This study suggested that neural cliques, as network-level functional coding units, were capable of overcoming the response variability of individual neurons and achieving real-time network representation for characterization and intensity information of startling episodic experiences.2. Dynamic activity of neural network in the hippocampal CA1 region encoding contextual and trace fear memoryTo further investigate the underlying mechanisms of associative fear learning and memory, we recorded the neural activity in the hippocampal CA1 region when animals underwent Pavlovian fear conditioning, and analyzed the data by using multiple discriminant analysis (MDA) and pairwise correlation analysis.Our results showed that ensemble neural responses to the tone increased after paring a tone (a conditioned stimulus, CS) with an electrical shock (an unconditional stimulus, US) compared with nearly no response before pairing, indicating that neural responses to the tone can be associated with neural responses to the US by pairing the CS with the US. The activated responses recurred when a tone was presented during the cued recall, which suggested that the associated responses can be retrieved by presenting appropriate cues. We also observed that sharp wave/ripple (SWR) occurring during freezing was similar with that during slow wave sleep periods (SWS). Pairwise firing-rate correlations within shock-responsive neurons increased during the freezing after training compared with that during exploration; whereas, correlations within non-responsive neurons did not change significantly. Correlations within responsive neurons also increased from pre-training SWS to post-training SWS; the increased correlations during post-training SWS were mostly confined to SWRs. These results suggested that the electrical-shock-induced neural co-activations could be reactivated for consolidation during freezing and could also be replayed during SWS, especially during ripple periods.Multiple discriminant analysis of ensemble neural activity delineated that the ellipsoid visualizing the neural responses to the tone in the 3-D MDA subspace, which was situated very close to the rest ellipsoid before training, moved towards the location representing the shock ellipsoid during pairing, and then was relocated to the new position which was further close to the shock ellipsoid after SWS, suggesting that the dynamical modification of neural network in the hippocampus could be triggered not only by external experiences, but also by internal reactivations of memory traces.This study investigated the activity of hippocampal neural network during the requisition, consolidation, and retrieval of fear memory, suggesting that ensemble neural activity dynamically modified the connecting weights of neural network for encoding the fear memory.3. Changes of synaptic weights through NMDA receptors modifying neural network in the hippocampal CA1 region during fear memoryWe have showed that ensemble neural activity in the hippocampus dynamically modified the connecting weights of neural network for encoding the fear memory, and we also have known that NMDA receptors were required for mediating synaptic plasticity. Therefore, we predicted that NMDA receptors in the hippocampus might play an important role in encoding fear memory through modulating synaptic plasticity. To address this question, we compared the ensemble neural activity in the hippocampal CA1 region in NMDA knockout mice with the control littermates.MDA analysis showed that startle-representative ellipsoids in 3-D subspace generated from NMDA knockout mice (KO mice) could be isolated from each other, suggesting that startle-related inputs could still be converged into CA1 region for adjusting the connectivity and specificity of the hippocampal network in KO mice despite the lack of NMDA-mediated plasticity; therefore, KO mice remained the capability of encoding startle episodes. Behavioral experiments showed that knockout mice failed to memorize the associations between conditioned stimuli (CS) and unconditioned stimuli (US). In vivo recording in KO mice showed that few shock-responsive neurons could associate responses to the US with the CS during training, suggesting that the impaired associations represented by individual neurons in the hippocampus of KO mice might directly interfere with fear learning. Pairwise correlation anaysis showed that there were not significant increases in between responsive neurons from pre-training slow wave sleep (SWS) session to post-training SWS, suggesting that impaired reactivation during the consolidation might be another main factor for the deficit in the fear memory. These findings indicated that MNDA receptors, as coincidence detectors, might contribute to the modification and maintenance of appropriate connectivity within memory-encoding neural networks during the requisition and consolidation of fear memory. Our study compared the ensemble neural activity in the hippocampal CAl region in NMDA knockout mice with the control littermates during encoding fear memory, indicating that MNDA receptors might be the key component for the acquisition and consolidation of fear memory.4. Direct control of machine by real-time memory codesWe have demonstrated the potential mechanisms in the hippocampus for encoding the fear memory. To apply these hypotheses, we built up a brain-computer interface by converting the activation patterns of these coding unit assemblies into sets of real-time digital codes for controlling physical devices. Two mathematical transformations, including neural clique method and multiple discriminent analysis method, were used to convert real-time ensemble neural activity into binary codes. These brain-derived binary codes were then decoded into commands for controlling physical devices, such as a mental diary recorder and a door controller. Neural activity in the hippocampal CAl regions of freely behaving mice in response to earthquake-like shaking stimuli was simultaneously recorded as signal sources to control physical devices. After continued training, the controlling signals could be also triggered by either neural responses to door closing, which were external cues associated with the shaking events; or reactivations of responsive neurons, which were internal signals without actual sensory inputs. The successful application of BCI suggested that startle-induced neural pattern changes could be used to control physical devices, which also confirmed the hypothesis that neural activity in the hippocampus participated in encoding the fear memory.Our studies investigated underlying mechanisms of fear learning and memory at the neural network level by comparing the ensemble neural activity in the hippocampal CAl region in NMDA knockout mice with the control littermates. Our progress in understanding neural coding, as well as technological advances, would contribute to further work in this field.