Don’t worry, it won’t be fine. Contributions of worry and anxious arousal to startle responses and event-related potentials in threat anticipation

A widely shared framework suggests that anxiety maps onto two dimensions: anxious apprehension and anxious arousal. Previous research linked individual differences in these dimensions to differential neural response patterns in neuropsychological, imaging, and physiological studies. Differential effects of the anxiety dimensions might contribute to inconsistencies in prior studies that examined neural processes underlying anxiety, such as hypersensitivity to unpredictable threat. We investigated the association between trait worry (as a key component of anxious apprehension), anxious arousal, and the neural processing of anticipated threat. From a large online community sample (N = 1,603), we invited 136 participants with converging and diverging worry and anxious arousal profiles into the laboratory. Participants underwent the NPU-threat test with alternating phases of unpredictable threat, predictable threat, and safety, while physiological responses (startle reflex and startle probe locked event-related potential components N1 and P3) were recorded. Worry was associated with increased startle responses to unpredictable threat and increased attentional allocation (P3) to startle probes in predictable threat anticipation. Anxious arousal was associated with increased startle and N1 in unpredictable threat anticipation. These results suggest that trait variations in the anxiety dimensions shape the dynamics of neural processing of threat. Specifically, trait worry seems to simultaneously increase automatic defensive preparation during unpredictable threat and increase attentional responding to threat-irrelevant stimuli during predictable threat anticipation. The current study highlights the utility of anxiety dimensions to understand how physiological responses during threat anticipation are altered in anxiety and supports that worry is associated with hypersensitivity to unpredictable, aversive contexts. Supplementary Information The online version contains supplementary material available at 10.3758/s13415-023-01094-4.


Supplement D: Full multilevel model for P3
To test the specificity of this effect to PSWQ we exploratorily included BDI-II into the model, which did not change the pattern of results. Specifically, the interaction (Predictable × PSWQ) remained significant (b = -0.03, SE = 0.01, t(789) = 2.03, p = .042).

Supplement E: t-transformed startle analyses
Of note, the random-intercept model failed to converge with t-transformed startle amplitudes.
The data were t-transformed within each individual, so the mean value for each participant was 50 (SD = 10). Hence, the t-transformation decreased the random variance in the data and thus a generalized linear model was estimated. 19.9 23, 4577 <.001 Note. For contrast coded conditions, neutral is the reference category, for contrast coded cues, interstimulus interval is the reference category. PSWQ = Penn-State Worry Questionnaire (grand-mean centered); MASQ = Mood and Anxiety Questionnaire (Anxious Arousal subscale, grand-mean centered); For contrast coded conditions, neutral is the reference category, for contrast coded cues, interstimulus interval is the reference category. PSWQ = Penn-State Worry Questionnaire (grand-mean centered); MASQ = Mood and Anxiety Questionnaire (Anxious Arousal subscale, grand-mean centered); σ 2 = residual variance; τ00 = random intercept; ICC = Intra class correlation; *** p < .001; ** p < .01; * p < .05;

Supplement G: Power Simulations
The power analysis was computed using simr version 1.0.5 (Green & Macleod, 2016).
Each analysis is based on 1,000 Monte Carlo simulations (Arend & Schäfer, 2019). For each outcome (i.e., startle, N1, P3), we simulated the power of the consistent highest order effect along the sample size. The resulting power estimates and power curves ( Figure S1) indicate the likelihood of finding comparable true effects in the data given the sample size of n = 101.
In the startle data, the consistent effect across analytical approaches was the cross-level interaction PSWQ × condition. According to the result of the power simulation, we could have detected comparable effects in the data with a likelihood of 90.10 % (95 % CI [88.08, 91.88]).
For N100, we simulated the level 1 effect of condition, as this was the only emergent effect in the data. The simulation result suggests that the power for finding comparable level 1 effects in the N100 data was 100 % (95 % CI [99.63, 100.0]).
For P300, we simulated the cross-level interaction of PSWQ × condition. The result indicates that comparable interaction effects in the P300 data could have been found with a probability of 70.10 % (95 % CI [67.16, 72.92]).

Figure S1
Power curves using 1,000 Monte Carlo simulations for the relevant effects of each outcome. Error bars depict the 95 % CI.

Figure S2
Grand averages and topographical depictions of the event-related potentials N1 and P3 collapsed across experimental conditions Note. The signal is locked to startle probes and is collapsed across conditions and cues of the NPU-Threat test. The time windows for quantification of N1 (± 25 ms area around the individual peak at FCz; white dot; ca. 120-170 ms) and P3 (mean activity 300 -370 ms at Pz; white dot) are indicated by shaded grey bars. The shaded area around the mean signal depicts ± 1 standard deviation of the distribution at each ms.