Cortical Oscillatory Dynamics Track Sympathetic Arousal and Index Individual Differences in Anxiety

Technical Summary: Cortical Oscillatory Dynamics Track Sympathetic Arousal and Index Individual Differences in Anxiety

Problem and Motivation
While hemodynamic imaging has identified a cortical network (including the anterior insula, anterior cingulate cortex, and default-mode network regions) central to autonomic regulation, its limited temporal resolution leaves critical ambiguities unresolved regarding the moment-to-moment orchestration of sympathetic arousal during threat. Specifically, the temporal ordering of cortical activity relative to sympathetic activation remains unclear, as does the identification of which spatiotemporal patterns track the magnitude of autonomic responses. Furthermore, direct evidence linking individual variability in cortical–autonomic coupling to the subjective experience of anxiety is lacking. This gap limits the understanding of why some individuals are more prone to anxiety than others, given that anxiety involves sustained sympathetic mobilization and that individuals differ markedly in how cortical activity covaries with autonomic responses.

Methodology
The study employed a multimodal approach combining magnetoencephalography (MEG) and electrodermal activity (EDA) in 85 healthy adults.

• Experimental Paradigm: Participants engaged in a sustained-threat "escape game" involving 14 sequences of visual scenes where they navigated a maze while being pursued by a predator. The task induced a graded build-up of anticipatory anxiety, with a 65% probability of receiving a mild electric shock. This design captures prolonged threat processing under uncertainty, differing from discrete cue–outcome paradigms.
• Data Acquisition: MEG data were recorded using a 306-channel Elekta TRIUX system, and EDA was recorded from the right foot. MEG data were source-localized to 448 cortical parcels using sLoreta and morphed to individual surfaces.
• Analytical Framework: A three-stage analytical pipeline was utilized:
  1. Deep Learning Decoding: A spatiotemporal convolutional neural network (LF-CNN) was trained to decode single-trial EDA response magnitudes from whole-brain MEG source activity preceding the autonomic response. This data-driven approach identified specific cortical regions and frequency bands predictive of sympathetic arousal without a priori constraints.
  2. Feature Extraction: Activation patterns from the trained model were analyzed to extract spatial (regions of interest) and spectral (frequency bands) features driving successful predictions.
  3. Confirmatory Modeling: Time-resolved linear mixed-effects modeling (LMEM) was applied to the data-driven regions and frequency bands. This quantified the precise spatial, spectral, and temporal contributions of oscillatory power to EDA magnitude while accounting for inter-individual variability through random effects. Subject-specific coupling slopes were derived and correlated with State-Trait Anxiety Inventory (STAI) scores (trait anxiety, pre-experiment state anxiety, and post-experiment state anxiety).

Key Results

• Predictive Accuracy: The LF-CNN model successfully decoded single-trial sympathetic response magnitude from MEG data, explaining 39.8% of the EDA variance on the held-out test set.
• Temporal-Spectral Dynamics: The model identified a temporally structured sequence of oscillatory dynamics preceding phasic EDA:
  • Early Phase (0.0–0.9 s): Characterized by beta-band (15–29 Hz) synchronization in the ventral visual stream (VVS) and posterior cingulate cortex (PCC), and gamma-band (31–39 Hz) synchronization in the right anterior insula.
  • Concurrent Early Phase: Theta-band (3–6 Hz) suppression in the right anterior insula.
  • Late Phase (1.1–1.6 s): Alpha-band (7–13 Hz) desynchronization in the anterior cingulate cortex (ACC).
• Individual Differences in Anxiety: Crucially, anxiety was not associated with mean oscillatory power or overall EDA amplitude alone. Instead, anxiety scores correlated with the strength of coupling between cortical oscillations and EDA:
  • Trait Anxiety: Negatively associated with beta-band coupling strength in the VVS. Higher trait anxiety corresponded to weaker co-variation between ventral visual beta power and sympathetic arousal.
  • State Anxiety (Post-Task): Associated with distinct patterns in the insula and PCC. Higher state anxiety correlated with stronger insular gamma–EDA coupling but attenuated (weaker) insular theta–EDA and PCC beta–EDA coupling.
  • Null Findings: Neither mean oscillatory power nor phasic EDA amplitude alone showed significant associations with trait or state anxiety.

Significance and Claims
The paper claims that sympathetic arousal during threat is not encoded by isolated regions or frequency bands but by the coordinated engagement of distributed cortical systems with a specific temporal structure. The primary contribution is the identification of neural–autonomic coupling strength as a distinct dimension of individual differences in anxiety.

The authors argue that anxiety should be understood not as a simple elevation in neural activity or physiological reactivity, but as a disruption in brain–body communication. Specifically:

• Trait anxiety appears linked to a decoupling of contextual, stabilizing signals (posterior beta) from autonomic output.
• State anxiety is characterized by a shift toward focal, salience-driven coupling (amplified insular gamma) alongside a loss of integrative control (attenuated insular theta and PCC beta coupling).

The study demonstrates that resolving the spatial, temporal, and spectral structure of neural–autonomic interactions is necessary to link fine-grained neurophysiological dynamics to the lived experience of anxiety. The findings suggest that the dynamic co-variation between neural and autonomic signals is the key metric for understanding individual vulnerability to anxiety, rather than the magnitude of either signal in isolation.

Limitations
The authors note that the observational design captures time-resolved associations rather than causal mechanisms. While neural oscillatory patterns temporally preceded sympathetic responses, the study cannot establish the direction of influence, interpreting findings strictly in terms of temporal covariation and coupling.