Abstract
Using contextual information to predict aversive events is a critical ability that protects from generalizing fear responses to safe contexts. Animal models have demonstrated the importance of spatial context representations within the hippocampal formation in contextualization of fear learning. The ventromedial prefrontal cortex (vmPFC) is known to play an important role in safety learning, possibly through the incorporation of context information. However, whether contextual representations are related to context-dependent expression of fear memory in humans remains unclear. Twenty-one healthy participants underwent functional MRI combined with a cue-context conditioning paradigm within a self-navigated virtual reality environment. We found robust differential skin conductance responses (SCRs) in the Threat context, but also within the Safe context, indicating fear generalization. Within the Safe context, vmPFC responses to the CS+ were larger than those in the Threat context. We furthermore found environment-specific representations for the contexts in the hippocampus to be related to fear expression and generalization. Participants with a weak context representation showed stronger fear generalization compared to participants with a strong context representation. Thus, a low neural representation of spatial context may explain overgeneralization of memory to safe contexts.
Introduction
A hallmark of anxiety and Post-Traumatic Stress Disorder (PTSD) is the "overgeneralization" of fear—feeling intense dread in a safe environment simply because it shares minor features with a dangerous one. While animal research has established the hippocampus as vital for spatial memory and the vmPFC as key for safety learning, the bridge between these two systems in humans has remained murky. This study sought to understand if the neural quality of a spatial "map"—how distinctly the brain represents one place versus another—predicts whether a person will correctly contextualize fear or let it bleed into safe environments.
Experimental Design and Methods
The study employed a sophisticated blend of behavioral conditioning, high-resolution fMRI, and machine learning. Twenty-one healthy volunteers participated in a paradigm consisting of four main phases: familiarization, cue localizer, context training, and fear conditioning.
Self-Navigated Virtual Reality (VR)
Participants volitionally navigated a 3D environment containing two buildings (Threat and Safe contexts). These buildings had distinct external features but identical interiors. Active movement was utilized to strengthen hippocampal "place cell" firing, providing a more ecologically valid measure of spatial representation than passive video viewing.
Contextual Fear Conditioning
Within each context, participants were presented with two visual cues: a Conditioned Stimulus (CS+) and a neutral stimulus (CS−). In the Threat building, the CS+ was consistently paired with an electric shock (100% reinforcement rate). In the Safe building, the same CS+ appeared but was never paired with a shock. The CS− was never paired with a shock in either environment.
Neural Decoding and Physiological Monitoring
Researchers used a Linear Support Vector Machine (SVM) classifier to analyze fMRI patterns in the hippocampus during the training phase, before any shocks were administered. This measured the inherent "strength" of a participant's spatial representation. Skin Conductance Responses (SCR) were recorded as a real-time proxy for fear levels.
Key Results
Fear Generalization and vmPFC Regulation
Participants showed robust differential SCRs (CS+ > CS−) in the Threat context. However, these responses generalized to the Safe context, confirming the paradigm successfully induced fear generalization. Crucially, vmPFC responses to the CS+ were significantly larger in the Safe context than in the Threat context. This suggests the vmPFC acts as a regulatory "brake," actively working to suppress fear output when the context is known to be safe.
The Quality of the Spatial Map
The SVM successfully decoded which building a participant was in with approximately 54% accuracy, significantly above chance. The most striking finding was that hippocampal "distinctness" measured before learning predicted future behavior. Participants with "weak" context representations (z-score < 1.65) exhibited significantly stronger fear generalization than those with "strong" representations (z-score > 1.65). If the hippocampus struggled to differentiate the two buildings during initial exploration, the individual was more likely to show high fear responses in the safe building later.
Discussion
These findings demonstrate that context-dependent regulation of fear expression engages ventromedial prefrontal pathways, suggesting a mechanistic overlap with the retrieval of extinction memory. The study highlights that the "sharpness" of the brain's internal spatial maps determines how well an individual can restrict fear to dangerous locations.
While the sample size of 21 participants is a limitation for individual difference analysis, the results are provocative. Future work should investigate these mechanisms in clinical populations, such as those with PTSD, to determine if hippocampal mapping deficits serve as a pre-existing risk factor for overgeneralization disorders. Furthermore, exploring non-spatial contextual factors (e.g., odors or sounds) could clarify which layers of environmental representation are most protective against maladaptive fear.
Summary updated