NMDA receptors involvement in memory and prediction

Memory and prediction are fundamental cognitive functions shaping our daily activities, with prediction enhancing the likelihood of successful action and relying heavily on memory. Studies indicate that spike timing-dependent plasticity (STDP), a mechanism altering synaptic weights based on temporal coordination, is crucial for memory, with NMDA activation playing a pivotal role. Our investigation aimed to assess the impact of inhibiting NMDA activation on long-term memory and prediction in random in-vitro networks of dissociated rat cortical neurons using MEAs.

During long-term memory formation, experiences are initially encoded in the hippocampus and later reactivated, triggering neocortical areas in systems consolidation. To study memory and prediction, we subjected cultures to focal, low-frequency electrical stimulation. For memory, a 10 min stimulation with 1 h intervals of spontaneous activity was employed. For prediction, 20 h focal stimulation with random interstimulus intervals was applied. NMDA receptors were blocked with amino-5-phosphonovaleric acid (APV).

Long-term memory formation was assessed  estimating functional connectivity with Conditional Firing Probabilities and possible connectivity changes using Euclidean distances. In control cultures, focal stimulation induced persistent connectivity changes, indicating memory trace formation. APV treated cultures showed no connectivity changes. Prediction and its relationship with memory were studied using mutual information between activity and future (prediction) or past (short-term memory) stimuli. Prediction occurred in both groups, but its dependence on short-term memory decreased over time only in the control group when long-term memory traces are formed.

In conclusion, our findings suggest that formation of long-term memory traces in random cortical networks critically depends on NMDA activation. Additionally, if long-term memory is impaired, prediction still requires short-term memory. Our in-vitro model using MEAs provides a simulation of the long-term memory process and prediction.