Cortical-Hippocampal Circuit Dynamics for Statistical Learning

New York University School Of Medicine, New York NY

Investigators

Abstract

PROJECT SUMMARY Statistical learning (SL) is a fundamental cognitive process that enables the brain to rapidly extract regularities from continuous experience. SL is a key component of many cognitive functions, for example, the ability to acquire words during language development. There has been a tremendous amount of behavioral research on SL, which is now accompanied by a growing neuroscience literature. These latter studies have revealed that both the cortex (e.g., superior temporal gyrus; inferior frontal gyrus) and the hippocampus are involved in SL. However, there exists no comprehensive account of what computations these regions perform and how these regions interact at a circuit level. Further, it remains unknown how SL emerges at increasing levels of complexity from rudimentary chunking based on transitional probabilities to the acquisition the rules of nested tree structures, such as in language. The broadest objective of our proposed work is to build and test a mechanistic neurobiological circuit theory of SL along the hierarchy of SL complexity. We hypothesize that SL occurs through dynamic interactions between the hippocampus and cortex. Testing this hypothesis requires a method that is both spatially precise and time resolved (with millisecond timing for quantifying neural dynamics). We will use non-invasive magnetoencephalography (MEG) in healthy participants alongside invasive electrophysiology in epilepsy patients, namely intracranial EEG (iEEG) with cortical surface electrodes and depth electrodes in the hippocampus. With high density hybrid depth electrodes, we will further enhance the spatial resolution in the hippocampus at the subfield level. To track the rapid evolution of statistical representations across brain regions, we will employ a novel neural frequency tagging (NFT) approach. NFT tracks SL by revealing brain rhythms at the frequency of regularities embedded in a continuous speech stream. Equipped with this cutting-edge toolkit, we will pursue our objectives across 3 aims that explore the mechanisms of SL at increasing levels of complexity. In Aim 1, we will examine functional coupling between the hippocampus and cortex during basic statistical learning of transitional probabilities using subfield resolution hippocampal recordings from iEEG and whole-brain source analysis in MEG. In Aim 2, we will examine how abstract rule-learning emerges in the brain. In Aim 3, we will investigate how SL unfolds during learning of hierarchical learning of nested structures. We hypothesize that learning specific vs. abstract regularities (i.e., rules and nested hierarchies) leads to differential engagement of the hippocampus and neocortex, respectively, resulting in reversal of information flow between hippocampus and neocortex. These findings will enhance our circuit-level understanding of SL, advancing the fields of language and learning, and highlighting potential new diagnostic and therapeutic avenues for comorbid cognitive deficits in neurological disorders.