Pattern separation and neurogenesis in the inhibitory network of adult rodent Dentate Gyrus

EMILIO KROPFF; JULIANA REVES SZEMERE

Congreso; ICBP 2017,; 2017

The dentate gyrus of the hippocampus is one of the few mammalian brain structures endowed with adult neurogenesis, a radical form of plasticity through which new neurons are inserted every day into pre-existing and fully functional networks. Although the last decades have witnessed significant progress in our understanding of the biological mechanisms that make this possible, and on the behavioral consequences of manipulating neurogenesis, we still lack ideas that vindicate the need of such a complex and expensive form of plasticity from the point of view of information processing. In other words, we do not know what special algorithm is being implemented in the dentate gyrus that cannot do without adult neurogenesis, which has been otherwise banished from almost the entire mammalian brain. Here we propose a simple competitive network that performs pattern separation (or decorrelation of inputs) by means of applying hebbian learning in a context of strong feedback inhibition. Both hebbian plasticity and inhibition are salient characteristics that differ in young compared to mature newborn neurons. We first study the fast dynamics of the network, describing analytically the equilibrium between feedforward excitation and feedback inhibition in response to a given input. Two parameters (the gain of the inhibitory circuit and the gain of excitatory newborn cells) control equally the amount of inhibition in the system, and thus the specificity of the response. We next study analytically the equilibrium of the slow hebbian learning dynamics in response to a sequence of inputs, finding a surprising dissociation in the role of these two parameters. While the gain of the feedback inhibitory circuits controls the number of neurons recruited by the learning process, the gain of excitatory newborn neurons controls the specificity of the response and thus pattern separation. This theoretical dissociation is also observed in network simulations. We finally discuss biological limitations to pattern separation by inhibition and predictions that could be tested experimentally to further deepen our understanding of the algorithm behind neurogenesis.