Memory networks in the human brain

CNBC Colloquium
Center for the Neural Basis of Cognition (CNBC)

Memory networks in the human brain

Michael Kahana, PhD
Professor, Department of Psychology and Director, Computational Memory Lab
University of Pennsylvania
September 20, 2018 - 4:30pm
328 Mellon Institute

Abstract: Human memory function is highly variable, fluctuating between periods of high and low performance even within a given person. Neurosurgical patients with indwelling electrodes present a unique opportunity to study the neural correlates of this variability and to define both the features of neural activity at a given brain location and the functional connections between brain regions that predict variability in memory encoding and retrieval. Here, I will describe our recent efforts to characterize brain networks that support memory via correlative (passive neural recording) and causal (direct electrical stimulation) approaches. Furthermore, many canonical memory regions emerge as hubs of such low- frequency connections, including the lateral frontotemporal cortices, the parahippocampal gyrus – and within it – the entorhinal cortex. High-frequency bands (i.e. gamma, 30+ Hz) almost exclusively exhibit desynchronization during successful memory operations. We recently extended these correlative studies and used intracranial stimulation to ask whether functional connections imply causality. We confirmed that electrical stimulation within the MTL evokes increases in theta power at remote regions, as predicted by the strength of low-frequency functional connections. However, this relationship held true only so long as stimulation occurred in or near white matter. These findings demonstrate the importance of low-frequency connectivity to episodic memory, integrating these findings over spatial scales and through causal and correlative approaches. Throughout the brain, we find that low-frequency networks exhibit reduced local power but stronger functional connectivity during successful episodic encoding and retrieval.