Decoding Fear or Safety and Approach or Avoidance by Brain-Wide Network Dynamics

Abstract

Discerning safety from threat and positive or negative outcomes of adversity are fundamental for mental health. Many brain structures have been implicated in both adaptive and maladaptive stress coping, however, how multiple regions function together as a network in the processing of this information is unclear. Here, we recorded local field potentials from seven regions of the mesolimbic-hippocampal-prefrontal cortical network (MLHFC) of male rats during the conditioning of a stimulus (CS) to the absence (safety) and then to the anticipation (fear) of footshocks, and during an approach-avoidance task. We developed a machine learning pipeline to investigate the relevance of specific features of oscillatory activity in the decoding of fear versus safety and approach versus avoidance. We found that decoding performance increased as a function of the number of brain regions included, reaching the best classification if all regions were considered. In addition, the best decoding was obtained from frequencies within the theta range (4-10 Hz). Remarkably, decoder models showed robust generalization within but not between individuals. Nevertheless, we were also able to identify patterns of MLHFC activity that decoded stress coping states from all rats. These patterns were characterized by increased brain-wide theta synchrony during fear and preceding approach. Our results indicate that stress coping information is encoded at the brain-wide level and highlight individual variability in this neural processing. Our findings also suggest that MLHFC network theta activity underlies active stress coping with both aversive and positive motivational valences.

Dysfunctional hippocampal-prefrontal network underlies a multidimensional neuropsychiatric phenotype following early-life seizure

“…PCA, using singular-value decomposition, was used for dimensionality reduction to find collective patterns of variance within behavioral variables (behavior factors) and neurobiological variables (neurobiology factors), and to reduce each immunohistochemical measure across all brain regions into single scores ( Johnson and Wichern, 2007 ; Marques et al, 2023 ; Marques et al, 2022 ). Principal components (PCs) were projected onto data and the mean score was compared between conditions.…”

“…To assess the effectiveness of neurophysiological features in discriminating CTRL from ELS rats, we fitted a regularized linear discriminant classifier model on the PCA scores of selected electrophysiological variables. We used the MATLAB function fitcdiscr without hyperparameter optimization ( Johnson and Wichern, 2007 ; Marques et al, 2023 ; Marques et al, 2022 ). The linear classifier using two predictors allowed the graphical representation and avoided the overfitting of a high number of variables.…”