Spike-timing dependent plasticity and synchronous oscillations in an invertebrate olfactory system

Abstract

Sensory systems neuroscience aims to study how patterns of neural activity represent stimuli of the outside world. To this end, the present work addresses how olfactory stimuli are represented by three successive layers in the locust olfactory system. Activation by an odorant of primary sensory neurons in the antenna gives rise to broadly distributed, oscillatory spatiotemporal activity patterns across the antennal lobe (AL). This is in marked contrast to the representation in the mushroom body (MB), where Kenyon cells (KCs) respond very sparsely and very briefly. In the AL, an odor gives rise to a particular trajectory through Projection Neuron (PN) phase space, with individual timepoints representing different aspects of the stimulus; in the MB, very small subsets of KCs respond selectively at particular timepoints along this trajectory. Two mechanisms are identified that contribute to the sparsening across the two structures: an intrinsic voltage dependence in the KCs, which gives rise to a superlinear response to synchronous inputs, and a canonical network motif, feedforward inhibition, which diminishes the KC response to nonsynchronous excitatory inputs. From a decoding perspective, this makes the oscillation cycle the relevant timestep of the AL trajectories, and it demonstrates a role for synchronous oscillations in sensory networks. While broad activation of the AL promotes extensive local interactions, giving rise to dynamic representations and enabling multiple features to be extracted, the sparse representation after decoding by KCs likely facilitates the storage of relevant patterns in memory. A subset of MB extrinsic neurons with dendrites densely invading the [beta]-lobe ([beta]LNs) is well placed to decode the KCs’ sparse responses. The synapses formed by KCs onto these cells are powerful and undergo Hebbian spike-timing dependent plasticity (STDP) on a timescale similar to the synchronous oscillations generated in the AL (and propagated through the MB). STDP has a homeostatic effect on the firing phase of [beta]LNs by fine-tuning the strength of KC-[beta]LN synapses, contributing to tight locking among subsets of [beta]LNs during odor stimulation and facilitating the flow of synchronous information. The facilitation of tight synchrony among [beta]LNs by STDP further ensures that different odor features computed and formatted as a function of cycle number by the AL, and represented by the sparse representations of KCs, remain segregated between LFP oscillation cycles. This segregation is also sustained by phase locked feedforward inhibition onto [beta]LNs, which restricts the window of integration for inputs from KCs, and is found to be due to neighboring [beta]LNs of the same class. The implications of the resultant competition among [beta]LNs due to this inhibition, and particularly its interaction with STDP at the KC-[beta]LN synapse are addressed with a network model. The results are considered within the context of the circuit in which the KC-[beta]LN network is embedded, and a cycle-specific mechanism for learning an arbitrary subset of the odor features computed in the AL is proposed.