Biophysically relevant network model of the piriform cortex predicts odor frequency encoding using network mechanisms

Abstract

Olfactory-guided animals utilize fast concentration fluctuations in turbulent odor plumes to perceive olfactory landscapes. Recent studies have demonstrated that the olfactory bulb (OB) encodes such temporal features present in natural odor stimuli. However, whether this temporal information is encoded in the piriform cortex (PCx) remains unknown. Hence, we developed a biophysically relevant PCx network model and simulated it using previously recorded in vivo activities of mitral and tufted cells in response to 2Hz and 20Hz odor frequencies for three stimulus mixtures. Analysis of single-cell activity across trials revealed that individual pyramidal neurons (PYRs) were largely ineffective at discriminating between 2Hz and 20Hz. However, the trial-averaged activity of the PYR population could discriminate between the two frequencies significantly. Moreover, using log-likelihood scores we further discovered that odor frequency discrimination happened through a highly distributed mechanism among the PYRs. One-dimensional convolutional neural network models trained and tested on PYRs activities achieved discrimination accuracies up to 95%. Using virtual synaptic knockout models, we found that eliminating either feedback or feedforward inhibition onto PYRs improved the decoding accuracy across all odor conditions. Conversely, eliminating recurrent excitation among PYRs or simultaneously eliminating recurrent inhibition within both interneuron populations degraded decoding performance. Removing recurrent connections within individual interneuron populations had minimal effects on the performance. Overall, our PCx model demonstrates that it can discriminate between 2Hz and 20Hz odor stimuli, with a bidirectional capability of performance modulation by specific circuit motifs. These findings predict that the piriform cortex encodes and processes temporal features of odor stimuli. New & NoteworthyWe simulated the first biophysically relevant network model of the piriform cortex (PCx) to show differential encoding of odor frequencies at 2Hz and 20Hz. 1D convolutional neural networks demonstrated a distributed role of pyramidal neurons in the encoding. Surprisingly, eliminating feedforward or feedback inhibition improves frequency discrimination, while eliminating recurrency impairs it. Specific circuit motifs, not just baseline activity levels, determine odor frequency representation. Overall, our results predict the PCxs capacity for temporal odor processing.