Brain imaging and machine learning reveal uncoupled functional network for contextual threat memory in long sepsis

Abstract

Abstract Positron emission tomography (PET) utilizes radiotracers like [18F]fluorodeoxyglucose (FDG) to measure brain activity in health and disease. Performing behavioral tasks between the FDG injection and the PET scan allows the FDG signal to reflect task-related brain networks. Building on this principle, we introduce an approach called behavioral task–associated PET (beta-PET) consisting of two scans: the first after a mouse is familiarized with a conditioning chamber, and the second upon recall of contextual threat. Associative threat conditioning occurs between scans. Beta-PET focuses on brain regions encoding threat memory (e.g., amygdala, prefrontal cortex) and contextual aspects (e.g., hippocampus, subiculum, entorhinal cortex). Our results show that beta-PET identifies a biologically defined network encoding contextual threat memory and its uncoupling in a mouse model of long sepsis. Moreover, machine learning algorithms (linear logistic regression) and ordinal trends analysis demonstrate that beta-PET robustly predicts the behavioral defense response and its breakdown during long sepsis.