Modeling Emergent Dynamics Arising from Synaptic Tagging and Capture at the Network Level

The transfer from short-term memory, which persists only for a few hours, to long-term memory is assumed to begin with a process called synaptic consolidation. Drawing from the hypothesis that memories are mainly stored by the strength of synaptic connections, synaptic consolidation means that a memory representation is consolidated by the stabilization of previous synaptic changes. The major candidate mechanism assumed to underlie this consolidation at the cellular level is synaptic tagging and capture (STC). Previous studies have provided the first pieces of evidence for this assumption considering simple feed-forward networks. However, the recurrent connectivity of hippocampal or cortical neuronal networks plays a particularly important role in memory function, as it enables pattern completion and temporally prolonged sequences of activity. Such dynamics mainly arise from so-called cell assemblies, which are groups of neurons with particularly strong synaptic connections. Thus, while it has been proposed that STC implements synaptic consolidation, the link between the physiological mechanisms of STC and the cognitive functions of long-term memory remains unclear. On timescales of minutes to hours (i.e., on the timescales of STC), these functions include memory improvement, selective consolidation, retroactive interference, and priming of a particular memory. Here, we first review findings from different computational studies of STC, and we present our computational model of STC-based synaptic consolidation in recurrent networks of spiking neurons. Then, we summarize the main results of our previous studies that suggest that STC can robustly implement the cognitive memory functions mentioned above. To this end, we have modeled the formation, consolidation, and improvement of memories represented by cell assemblies. Furthermore, we have shown that when the synthesis of plasticity-related proteins or products depends on neuromodulation, the level of neuromodulator can retroactively control the storing of different types of information. Moreover, we could demonstrate that emergent effects arise from the influence of STC on the interaction of multiple memory representations in different organizational paradigms. Through these findings, we provide a mechanistic explanation and contribute further evidence that STC plays an important role in various phenomena related to long-term memory. Finally, we discuss future steps for enabling computational models to predict the outcome of STC-related experiments in greater detail.