Modelling short and long-term synaptic plasticity in neocortical microcircuits

Learning and memory storage is believed to occur at the synaptic connections between neurons. During the last decades it has become clear that synapses are plastic at short and long time scales. Furthermore, experiments have shown that short and long-term synaptic plasticity interact. It remains unclear, however, how is this interaction implemented and how does it impact information processing and learning in cortical networks. In this thesis I present results on the mechanisms and function of this interaction. On the mechanistic level this form of plasticity is known to rely on a presynaptic coincidence mechanism, which requires the activation of presynaptic NMDA receptors (preNMDARs). In a collaborative effort I used mathematical modeling combined with experiments to show that preNMDARs reroute information flow in local circuits during high-frequency firing, by specifically impacting frequency-dependent disynaptic inhibition mediated by Martinotti cells. In order to accurately characterize how do preNMDARs regulate the release machinery, I developed a probabilistic inference framework that provides a distribution over the relevant parameter space, rather than simple point estimates. This approach allowed me to propose better experimental protocols for short-term plasticity inference and to reveal connection-specific synaptic dynamics in the layer-5 canonical microcircuit. This framework was then extended to infer short-term plasticity from preNMDAR pharmacological blockade data. The results show that preNMDARs up-regulate the baseline release probability and the depression time-constant, which is consistent with experimental analysis and that their impact appears to be connection-specific. I also show that a preNMDAR phenomenological model captures the frequency-dependence activation of preNMDARs. Furthermore, preNMDARs increase the signal-to-noise ratio of synaptic responses. These results show that preNMDARs specifically up-regulate high frequency synaptic information transmission. Finally, I introduce a pre- and postsynaptic unified mathematical model of spike-timing- dependent synaptic plasticity. I show that this unified model captures a wide range of short-term and long-term synaptic plasticity data. Functionally, I demonstrate that this segregation into pre- and postsynaptic factors explains some observations on receptive field development and enable rapid relearning of previously stored information, in keeping with Ebbinghaus’s memory savings theory