Pattern completion, or the ability to retrieve stable neural activity patterns from noisy or partial cues, is a fundamental feature of memory. Theoretical studies indicate that recurrently connected auto-associative or discrete attractor networks can perform this process. Although pattern completion and attractor dynamics have been observed in various recurrent neural circuits, the role recurrent circuitry plays in implementing these processes remains unclear. In recordings from head-fixed mice, we found that odor responses in olfactory bulb degrade under ketamine/ xylazine anesthesia, while responses immediately downstream, in piriform cortex, remain robust. Recurrent connections are required to stabilize cortical odor representations across states. Moreover, piriform odor representations exhibit attractor dynamics, both within and across trials, and these are also abolished when recurrent circuitry is eliminated. Here, we present converging evidence that recurrently-connected piriform populations stabilize sensory representations in response to degraded inputs, consistent with an auto-associative function for piriform cortex supported by recurrent circuitry.

These are processed and spike sorted recordings from olfactory bulb and piriform cortex under awake and anesthetized conditions. Documentation on the format of the files in this data can be found with a related dataset at https://crcns.org/data-sets/pcx/pcx-1/about-pcx-1

MATLAB scripts that use this data to produce the figures in the associated paper can be found at https://github.com/FranksLab/eLife2020-recurrents-stabilize