Dynamic and selective engrams emerge with memory consolidation

The cellular basis of episodic memory involves experience-activated neuronal ensembles, termed engrams, which are both necessary and sufficient for memory recall. However, the dynamic changes in engram composition and selectivity following initial encoding remain unclear. Two competing hypotheses exist: (1) engrams stabilize during consolidation, maintaining a consistent neuronal composition; (2) engrams are dynamic, with neurons joining and leaving the engram over time. The latter hypothesis is supported by observations of low overlap between neuronal ensembles activated during learning and recall (10-40%). Understanding the temporal dynamics of engrams is crucial to elucidate the relationship between engram composition and mnemonic properties, such as memory selectivity – a key element of adaptive behavior. This study employs both computational modeling and experimental approaches in mice to investigate the post-encoding evolution of memory engrams and their relationship to memory selectivity.

A significant body of research indicates the central role of experience-activated neurons in memory. Loss-of-function studies demonstrate that ablating these neurons disrupts memory retrieval, while gain-of-function studies show that artificial reactivation of these neurons triggers memory recall even without retrieval cues. These learning-activated neurons form the engram cell population. However, the stability of these engrams after initial encoding remains a key area of investigation. Previous research presents contrasting perspectives on engram evolution during consolidation, proposing either stable or dynamic engrams. The understanding of engram dynamics, particularly in relation to memory selectivity, is crucial for comprehending adaptive behavior. Recent studies highlight the emergence of memory selectivity in conditioned taste aversion over a timescale of hours, providing a relevant context for this research. Existing models focusing on stable memory traces have provided valuable insights into network dynamics during memory formation and recall. This study aims to build upon this foundation to understand the dynamic nature of memory engrams and their interaction with memory expression.

This study combined computational modeling with experimental techniques in mice. A spiking neural network model was developed, incorporating feedforward and recurrent excitatory synapses with short-term and long-term plasticity, and inhibitory synapses with inhibitory plasticity. Long-term excitatory synaptic plasticity integrated Hebbian (triplet spike-timing-dependent plasticity - STDP) and non-Hebbian (heterosynaptic and transmitter-induced plasticity) mechanisms. Inhibitory synaptic plasticity was activity-based STDP, regulating network activity levels. The model was trained using a simulated episodic memory task, and engram cells were identified based on selective activation to the training stimulus. A consolidation phase, involving reactivation of the training stimulus, followed, with engram cell status assessed at intervals. The recall phase tested selectivity using partial cues of the training stimulus and novel stimuli. Experiments used contextual fear conditioning (CFC) in mice. The Cal-Light system, employing three adeno-associated viruses (AAVs), was used for activity-dependent labeling of neurons in the dentate gyrus (DG). Blue light tagged active neurons during fear training (EGFP), and c-Fos staining identified recall-activated neurons. Optogenetic reactivation of Cal-Light-labeled neurons was used to assess sufficiency for recall. Longitudinal calcium imaging in the DG tracked neuronal activity across the CFC protocol. Optogenetic and chemogenetic manipulations were used to investigate the role of inhibitory interneurons (specifically CCK+ and PV+) in memory selectivity and plasticity. Slice recordings measured plasticity of CCK interneuron synapses onto labeled neurons. Statistical analysis included Wilcoxon signed-rank tests, one-sample Wilcoxon signed-rank tests, one-way ANOVAs, paired t-tests, and Spearman’s rank correlations, with significance set at P < 0.05. Non-negative matrix factorization was used in some analysis to identify engram cells.

The computational model predicted and the experiments confirmed that: 1. **Engrams are dynamic:** Neurons are continuously added to and removed from the engram during consolidation. The fraction of training-activated engram cells remaining in the engram decreased significantly over 24 hours, while new neurons became incorporated into the engram. This dynamic turnover was observed both in simulations and in experimental data using Cal-Light labeling and c-Fos staining in mice. 2. **Engrams transition from unselective to selective:** Initially, memory recall was similar for training and novel contexts. However, memory selectivity emerged after 12 hours of consolidation, as evidenced by differential freezing behavior in the training and neutral contexts in mice. This was consistent with the model's prediction that engrams become selective over time. 3. **Inhibitory activity is essential for memory selectivity:** Blocking inhibitory neurons during recall impaired memory selectivity in simulations. In mice, optogenetic inhibition of DG CCK+ interneurons during recall abolished memory discrimination, while inhibiting PV+ interneurons had no effect, reflecting the experimental findings. 4. **Inhibitory synaptic plasticity shapes selective engrams:** Blocking inhibitory synaptic plasticity during consolidation impaired selectivity in the model. In mice, chemogenetic inhibition of DG CCK+ interneurons immediately after training prevented the development of memory selectivity. Slice recording experiments showed this chemogenetic manipulation also blocked the plasticity of CCK+ efferent synapses onto training-activated engram cells. The longitudinal calcium imaging experiments showed a correlation between the dynamic engram activity and the freezing discrimination ability.

This study provides strong evidence that memory engrams are dynamic structures undergoing substantial reorganization during consolidation, a process crucial for the emergence of memory selectivity. The model successfully integrated the seemingly contradictory observations that training-activated neurons remain necessary and sufficient for memory recall despite considerable engram turnover. The findings challenge classical theories of static memory traces, highlighting the importance of ongoing synaptic plasticity in shaping memory representations. The crucial role of inhibition and inhibitory synaptic plasticity in both the expression and development of memory selectivity was demonstrated. The specific involvement of CCK+ interneurons, but not PV+ interneurons, underscores the importance of cell-type-specific inhibitory mechanisms in memory processing. This work has implications for understanding conditions characterized by persistent unselective aversive memories, such as PTSD and panic disorders.

This study demonstrates that memory engrams are dynamic and that their continuous reorganization during consolidation is crucial for the development of memory selectivity. The interplay between engram dynamics, inhibitory plasticity, and memory expression was clarified. Future research could explore the specific molecular mechanisms underlying inhibitory plasticity in engram cells and its modulation by sleep or other factors. Investigating the role of engram dynamics in other memory systems and exploring the link between engram plasticity and pathological conditions like PTSD would also be valuable.

The mouse model may not fully capture the complexity of human memory. The sample size in some experiments, particularly the longitudinal calcium imaging, was relatively small, warranting further investigation with larger cohorts. The reliance on indirect measures of neuronal activity (c-Fos, Cal-Light) in some experiments limits precise quantification of neuronal responses.