Data-Constrained Recurrent Network Neural Model Uncovers the Circuit Mechanism of Olfactory OFF Responses.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Insect Olfaction

Background:

• Sensory neural circuits must encode stimulus presence and termination.
• Projection neurons (PNs) in the insect antennal lobe (AL) exhibit OFF responses after odor offset.
• Circuit mechanisms generating these OFF responses in recurrent excitatory-inhibitory networks are poorly understood.

Purpose of the Study:

• To investigate the circuit mechanisms underlying OFF responses in the locust AL.
• To reconstruct odor-evoked temporal dynamics of PNs using a biologically-constrained recurrent neural network (RNN).
• To identify the pathways and network interactions responsible for generating OFF responses.

Main Methods:

• Constructed a biologically-constrained firing rate-based RNN model of the locust AL.
• Trained the RNN on 110 in vivo PN electrophysiological recordings across five odorants.
• Used targeted input and connectivity perturbations within the RNN to dissect circuit mechanisms.

Main Results:

• The trained RNN model accurately reproduced in vivo PN firing rates and temporal dynamics.
• OFF responses were found to arise from two distinct pathways: feedforward ORN input and a recurrent pathway.
• The recurrent pathway predominantly involves LN-LN mutual inhibition, generating net excitation through transient inhibition release.
• Two pathways recruit largely non-overlapping PN populations, suggesting OFF response identity is an emergent network property.

Conclusions:

• OFF responses in the AL are generated by a combination of feedforward and recurrent network mechanisms.
• Recurrent excitation for OFF responses is primarily mediated by the transient release of inhibition via LN-LN mutual inhibition.
• OFF response specificity arises from network state rather than cell-intrinsic features.
• Data-constrained RNNs are effective tools for dissecting neural circuit mechanisms from in vivo recordings.