Leveraging neural decoding to uncover biases in latent cause inference during fear extinction a core feature of anxiety disorders is persistent fear even after the threat has dissipated

The National Institute of Mental Health awarded The Trustees of Princeton University $152,600 under the Mental Health Research Grants program (CFDA 93.242) on November 1, 2026, to conduct neural decoding research on biases in latent cause inference during fear extinction in anxiety disorders.

The research leverages computational frameworks and neural decoding from optically-pumped magnetoencephalography (OPM-MEG) in combination with physiological markers including skin conductance response and pupillometry to decode the contents of inferred cognitive states and identify neural fingerprints of distinct pathways to sustained fear following extinction learning. The project investigates whether persistent threat responses in anxiety disorders arise from overgeneralization of threat across contexts—specifically, whether individuals infer a single latent cause combining both safe and threatening cues that remains persistently linked to threat, as opposed to successfully extinguishing the threat association.

Work is performed at Princeton, New Jersey, with a period of performance from November 1, 2026, through October 31, 2029. This is a project grant, a standard assistance type under the Mental Health Research Grants program supporting competitive research proposals evaluated on scientific and technical merit, program relevance, and program balance through peer review and the National Advisory Mental Health Council.

Here, we propose to leverage the computational framework of latent cause inference-which posits that people differ in the extent to which they assign observed events to distinct underlying contexts (or causes)-to uncover separable pathways to a sustained threat response. While previous evidence suggests that higher anxiety is related to inferring two separable latent causes-one safe and one threatening-and preferentially reinstating the threatening one, this partitioning is at odds with research linking pathological anxiety to the over-generalization of threat across contexts. Consistent with the generalization account, the sustained fear response could instead arise from inferring a single latent cause that combines both safe and threatening cues and is persistently linked to threat. Importantly, the contents of this inferred latent cause would be different from those of a single latent cause in which the association between a given stimulus and threat has been successfully overridden through extinction. While previous work has assessed whether extinction behavior is better described by models that assume one or more latent causes, IT does not provide information about the observations these causes contain. Thus, we do not yet know whether a persistent threat response can indeed be explained by the generalization of threat from acquisition to extinction. Here, we propose to leverage well-established markers of fear conditioning (skin conductance response and pupillometry) in combination with neural decoding from OPM-MEG to decode the contents of the inferred states and identify reliable "fingerprints" of these different pathways to sustained fear after extinction. While the physiological data will serve as an implicit measure of threat response, neural decoding will allow US to detect the contents of the latent cause participants inferred, allowing US to make the crucial distinction between a generalized threat state and a learned safe state. In aim 1, we will test whether individual differences in anxious symptomatology map onto these distinct pathways to sustained fear. We hypothesize that people high in anxious apprehension, which is linked to worry and rumination, will infer multiple causes and show return of fear due to the selective maintenance of the threatening latent cause. Conversely, we predict that people high in anxious arousal, which is characterized by sustained hypervigilance, will tend to integrate acquisition and extinction into a single latent cause that is predictive of threat. Aim 2 will examine the developmental trajectory of fear learning and extinction. We predict that while children will successfully unlearn the threat response, adolescents will infer a single latent cause associated with the aversive outcome, leading to a persistent threat response. Together, these studies will reveal meaningful heterogeneity in the mechanisms supporting sustained fear, paving the way for personalized therapeutic interventions across both symptoms and the lifespan.