4 Hz Theta Power Underlies Memory Encoding in Auditory Cortex in an Unsupervised Sequence Learning Paradigm

Abstract

Perceiving sensory information and integrating it with memory is a fundamental component of cognition, one that enables prediction-making. In mouse auditory cortex, it has yet to be shown if and how neural oscillations might support the encoding of sensory, memory, and predictive information. To test this, I analyzed LFP recorded from mouse auditory cortex in an implicit sequences learning paradigm embedded with statistical regularities. After performing a continuous wavelet transform to decompose the signal into time-frequency space, I trained linear support vector machine (SVM) classifiers to decode sensory, memory, and predictive information from the LFP signal and the power of neural oscillations at specific frequencies. I found that sensory and memory information are represented differently from each other in frequency-space: whereas sensory information was found widely distributed across multiple frequencies, memory encoding was distinctly localized in the low theta (∼4 Hz) and high gamma (∼90 Hz) frequency bands. Low theta power was sustained, stably representing sensory information and perhaps corresponding to top-down entrainment from hippocampus to support memory consolidation. Here, it is also shown that predictions and postdiction --- retroactive updates to sensory percepts in accordance with new stimuli --- were both facilitated by an alignment in the representation of correlated stimuli in time-frequency space. Future work should explore if phase-dependent encoding disambiguates sensory and memory information within frequencies in the LFP power. It also remains unclear if gamma amplitude is coupled to the phase of auditory or even hippocampal theta oscillations, yielding another promising area of research.