ABSTRACTPattern separation is a fundamental brain computation that converts small differences in synaptic input patterns into large differences in action potential (AP) output patterns. Pattern separation plays a key role in the dentate gyrus, enabling the efficient storage and recall of memories in downstream hippocampal CA3 networks. Several mechanisms for pattern separation have been proposed, including expansion of coding space, sparsification of neuronal activity, and simple thresholding mechanisms. Alternatively, a winner-takes-all mechanism, in which the most excited cells inhibit all less-excited cells by lateral inhibition, might be involved. Although such a mechanism is computationally powerful, it remains unclear whether it operates in biological networks. Here, we develop a full-scale network model of the dentate gyrus, comprised of granule cells (GCs) and parvalbumin+ (PV+) inhibitory interneurons, based on experimentally determined biophysical cellular properties and synaptic connectivity rules. Our results demonstrate that a biologically realistic principal neuron–interneuron (PN–IN) network model is a highly efficient pattern separator. Mechanistic dissection in the model revealed that a winner-takes-all mechanism by lateral inhibition plays a crucial role in pattern separation. Furthermore, both fast signaling properties of PV+ interneurons and focal GC–interneuron connectivity are essential for efficient pattern separation. Thus, PV+ interneurons are not only involved in basic microcircuit functions, but also contribute to higher-order computations in neuronal networks, such as pattern separation.
A neural network model of hippocampal contributions to category learning
