Persistent fear is a hallmark of anxiety disorders. This longitudinal naturalistic fMRI study examined whether persistent fear relates more closely to the cross-phase consistency of neural processing linked to conditioned threat or to the reexpression of neural patterns tied to subjective fear. The investigation compared two independently developed, pre-trained neural signatures—one representing subjective fear and one representing threat conditioning—to test associations with individual differences in trait anxiety.
Forty-two participants completed two separate fMRI phases spaced by approximately three months. In each phase participants viewed the same set of 18 naturalistic videos, followed by a resting-state segment. Combining video and rest segments produced 19 analysis segments per phase for cross-phase comparisons. The study therefore examined within-subject consistency in signature expression when participants encountered the identical naturalistic stimuli at two time points.
The authors applied two pre-trained neural signatures to each participant’s fMRI data to estimate expression strength over time. One signature was developed to predict subjective fear experiences; the other was trained to index threat-conditioning responses. Each signature was developed by a separate research group prior to this study. The approach treated expression time series of each signature as predictors of within-subject cross-phase stability and allowed testing whether higher trait anxiety related to more consistent reexpression across phases.
Across participants, higher trait anxiety was significantly associated with greater cross-phase consistency in the expression of the subjective-fear signature. In contrast, there was no significant association between trait anxiety and cross-phase consistency of the threat-conditioning signature. These results indicate that individuals with higher trait anxiety more consistently reexpress distributed neural patterns linked to subjective fear across repeated exposures to the same naturalistic contexts.
The observed association between trait anxiety and subjective-fear signature consistency became more prominent when analyses used longer temporal windows, suggesting that temporal aggregation increased the detectable relationship. By contrast, analyses of average regional time series from canonical fear-related regions—the amygdala, ventromedial prefrontal cortex (vmPFC), insula, and hippocampus—did not reveal the same association with trait anxiety. This dissociation implies that distributed multivariate pattern signatures may capture individual differences that are not detectable with single-region average activity measures.
The findings support the view that individual differences in trait anxiety are more closely tied to the stable reexpression of distributed neural patterns reflecting subjective fear than to the consistency of neural signatures associated with conditioned threat response. Methodologically, the study illustrates how applying previously validated predictive signatures to longitudinal, naturalistic fMRI data can reveal trait-linked consistency in neural pattern expression across repeated real-world-like experiences. Clinically, the results suggest that multivariate predictive models of subjective emotional states might be useful for characterizing trait-level vulnerability in anxiety.
The abstract provides key design features and primary results but does not report several methodological details. The summary does not list participant demographics beyond sample size, specifics of trait-anxiety measurement (scale used, distribution), statistical effect sizes or thresholds, preprocessing steps, motion control procedures, or correction for multiple comparisons. Information on whether findings were replicated in independent samples or cross-validated within the dataset is also not reported in the abstract. For these details readers should consult the full preprint.
In a repeated naturalistic viewing paradigm, higher trait anxiety predicted greater cross-phase consistency in a distributed subjective-fear neural signature but not in a threat-conditioning signature or in mean regional signals from classic fear-related areas. These results highlight the value of combining predictive neural models with longitudinal naturalistic designs to probe stable individual differences in emotional processing. The full preprint should be consulted for complete methods, statistics, and supplementary analyses.