Place Field Dynamics as a Window on Synaptic Plasticity in the Hippocampus

Although changes in synaptic efficacy are widely thought to be a substrate for memory storage in the brain, the rules of synaptic plasticity are difficult to assess in vivo and thus remain unclear. We considered the dynamics of hippocampal representations as an indirect indicator of ongoing plasticity during memory formation and familiarization. Using calcium imaging data from mice navigating virtual tracks, during which CA1 or CA3 neurons were recorded, we tracked the evolution of the spatial modulation of their activity (place fields) lap by lap. During exploration of a novel environment, 50% of CA1 place fields shifted, mostly backwards. The shift speed and proportion of shifting place fields was lower in CA3. To determine the mechanisms supporting these shifting dynamics, we simulated spiking place cells with plastic synapses following different plasticity rules. In contrast to early models, we found that classic Hebbian spike-timing-dependent-plasticity (STDP) is ill-suited to produce strong backward shifting when realistic firing rates and place field properties are maintained. Including inputs shifting like observed in CA3 was not sufficient to explain the shifts in CA1. Behavioral Timescale Synaptic Plasticity (BTSP), recently discovered in CA1, is an alternative candidate mechanism. Our new model of BTSP combined with homeostatic synaptic normalization points to BTSP, not STDP, as the main plasticity mechanism underlying the dynamics of spatial representations in CA1. It also suggests that BTSP occurs in CA3, albeit with phenomenological differences. Exploration of our model’s parameter space and mining of the experimental data for BTSP signatures further shows that the probability of BTSP-triggering events is dynamic: it decreases after place field onset and with familiarity to the environment. Overall, our study supports BTSP as a prominent type of plasticity in vivo, provides new insights in its phenomenology and new ways to incorporate it in computational models of spiking neurons.