Participants
Following the sample size in the prior tACS research targeting on prefrontal cortex (Alexander et al., 2019), we aimed to recruit 30 valid participants in each of the three groups (i.e., alpha, gamma and sham groups). To account for potential dropouts and attrition, we recruited a total of 110 participants from a local university. Eligibility was determined through an online pre-screening process, where participant completed the Beck Depression Inventory-II (BDI-II; Beck et al., 1996) and answered questions related to the tACS implementation. To be included in the experiment, participants’ BDI-II score should be lower than 29, without current or historical diagnoses of psychiatric disorders, without electronic implants, no history of head injury, and normal or corrected-to-normal vision. Subsequent exclusions were based on the following criteria: suspicion regarding the veracity of feedback probability (n = 7), quit the experiment (n = 3), falling asleep during the stimulation (n = 1), poor EEG data quality (n = 6). After these exclusions, the final sample included 31 participants in the alpha-tACS group (18 males, age: M ± SD = 20.52 ± 1.48), 31 participants in the 40 Hz-tACS group (12 males, age: M ± SD = 20.45 ± 1.48), and 31 participants in the sham control group (16 males, age: M ± SD = 20.66 ± 1.94). Participants received monetary compensation for their time (250 CNY or ~ 36 USD). All participants provided written consent before participation. The Human Research Ethics Committee of Shenzhen University approved the study.
Experimental Design and Task Procedures
Participants visited the lab on two consecutive days. On Day 1, participants completed the baseline belief update task to assess optimism bias, received the tACS, and the post-tACS assessments. After 24 hours, participants came back to the lab for a delay test on Day 2. An overview of the timeline of experimental tasks is illustrated in Fig. 1A.
Day 1 Lab session
Pre-tACS phase: Participants were given task instructions and four practice trials to familiarize with the belief update task. For baseline belief update task, participants completed a total of 48 trials, separated in two blocks with a one-minute break in between. Following baseline optimism bias assessment, 2-minute resting state EEG data were collected to determine each participant’s individual alpha frequency (IAF). During the resting-state, participants were instructed to maintain relax and keep their eyes open while fixating on the center of the screen. Individual stimulation intensity was determined during calculation of the IAF.
TACS phase: Subsequent to the determination of IAF and stimulation intensity, the stimulation protocol began, which lasted for approximately 15 minutes (see details below in tACS protocol). During the tACS stimulation, participants completed a modified mental rotation task (Kasten & Herrmann (2017). By engaging in this mental rotation task, we aimed to reduce the individual variances of the brain and cognitive states during the tACS stimulation (Ruhnau et al., 2016).
Post-tACS phase: Following the stimulation, participants spent two minutes for resting-state EEG to calculate the post-stimulation resting FAA. Participants then completed the tACS side-effect questionnaire. Participants next completed the post-tACS belief update task, with a total of 48 trials separated in two blocks.
Additional assessments: A surprise cued recall task was introduced following the post-tACS belief update task. Here, prompted by each adverse life event from the previous belief update task, participants shall recall the feedback probability shown earlier. Similar to Yao et al., (2021), this cued recall task assessed participants’ memory accuracy of the feedback. Finally, participants completed three questionnaires to assess trait anxiety (State-Trait Anxiety Inventory, STAI-T; Spielberger, 1983), depression symptoms (Beck Depression Inventory-II, BDI-II; Beck et al., 1996), and trait optimism (Life Orientation Test-Revised, LOT-R; Scheier et al., 1994).
Day Two Lab session
Approximately 24 hours after the Day 1 session, participants returned to the lab for the second lab session on Day 2. They then engaged in the following tasks in order: (1) a surprise belief update task to assess delayed optimism bias; (2) a surprise cued recall task to assess long-term memories for feedback probability; (3) a rating task involving the assessment of adverse life events on a 6-point Likert scale; and (4) additional questionnaires (including the STAI-T, BDI-II, and LOT-R).
Upon completion of all tasks, participants were debriefed and compensated accordingly.
Experimental Tasks
The experimental tasks, implemented in E-Prime® 3.0 (Psychology Software Tools, Inc., Sharpsburg, Pennsylvania, USA), are detailed as follows:
Belief update task (Fig. 1B). This task was adapted from Yao et al., (2021). Specifically, each trial presented an adverse life event for 2 s, after which participants had 8 s to estimate, and to type down the likelihood of themselves experiencing the event in their future lifetime (first estimation, E1). Following a blank interval (1.2–1.5 s), the feedback probability was shown for 2 s. Following another blank interval (1–1.2 s), participants had 2 s to re-evaluate their prior estimation, and to provide their second estimation (E2) within the next 8 s. Participants were informed that the feedback probability for each adverse life event was obtained from a large-scale study with a demographically similar population. Estimation error (EE) was defined as the differences between feedback probability and participants’ first estimation (E1 minus feedback). A trial was classified as desirable (vs. undesirable) when feedback probability was smaller (vs. larger) than E1. Immediate belief updating was calculated as the differences between participants’ second and first estimation (i.e., E1-E2 for desirable; E2-E1 for undesirable).
Recall Task. Participants were instructed to recall the previously presented feedback probability, prompted by each adverse life event. Each trial began with an adverse life event presentation (2 s) from the earlier belief update task, followed by a question mark (1s). Participants were given 8s to type down their remembered feedback probability. The Memory error was calculated as the absolute difference between recalled probability and presented feedback probability, i.e., |feedback-recall|, with larger values indicating higher memory errors.
Day 2 Estimation Task. Participants were presented with each of the 96 adverse life events from Day 1 and gave their third estimation (E3) without the presentation of any feedback. Delayed belief updating was calculated as the differences between participants’ third and first estimation (i.e., E1-E3 for desirable; E3-E1 for undesirable).
The participant-level optimism updating bias was calculated using immediate/delayed desirable updating minus immediate/delayed undesirable updating, with larger values indicating higher updating for desirable than for undesirable feedback, i.e., optimism biases.
Simplified Mental Rotation Task. This task was adapted from previous research, and was used to minimize individual differences in mental states during tACS (Kasten et al., 2019). During the task, a total of ten uppercase letters, with each letter repeating twice, would be presented in either normal or mirrored orientation. The inter-trial interval (ITI) randomly varied between 10 to110 seconds. Participants had to discern the orientation of the letter within a maximum of 5 seconds by pressing the space bar. Participants were instructed to relax while focusing on the screen to ensure timely responses to the letters.
Assessment of adverse life events. Participants rated each life event on a 6-point Likert scale across four dimensions: personal relevance, familiarity, vividness, and prior experience.
Details on the estimation error (EE) manipulation and event rating are presented in the supplement materials.
Electroencephalography (EEG) Data Acquisition
EEG data were recorded using a Brain Products Brain Vision Recorder (Brain Products GmbH, Munich, Germany) with 64 active Ag/AgCl electrodes mounted in a cap and located in the standard positions according to the International 10–10 system. The ground electrode was positioned at AFz, the online reference was positioned at FCz, and the electrode impedance was kept below 20 KΩ. The sampling rate was 1000 Hz using a BrainAmp AC amplifier (Brain Products GmbH, Munich, Germany).
Transcranial alternating current stimulation (tACS) protocol
Randomization
Participants were randomly assigned to one of the three groups (IAF-tACS, 40 Hz-tACS, or sham), with no more than two consecutive similar assignments. After recruiting approximately 20 participants per group, a gender imbalance emerged, with only five males in the 40 Hz-tACS group and six in the IAF-tACS group. To address this, male participants were randomly assigned to the sham or 40 Hz-tACS groups, while female participants to the IAF-tACS or sham group until an equal gender distribution was achieved across the groups. Subsequently, random allocation resumed. Group allocation remained blinded to the participants.
Determination of IAF and Individual Current Intensity
The two-minute eyes-open resting-state EEG data collected prior to the stimulation was analyzed to determine the IAF for the IAF-tACS group. Artifact-free EEG segments were identified through visual inspection and further cleaned using independent component analysis (ICA) with the ICLabel plugin (v1.2.6, Pion-Tonachini et al., 2019) in EEGLAB (Delorme & Makeig, 2004). The IAF calculation was performed on the clean EEG data from frontal and parietal electrodes (F4, F3, Pz, and an averaged across a right frontal electrodes cluster comprising F4, F2, FC4, F6, and AF4). The Fast Fourier Transform (FFT) method was utilized to obtain the IAF. For the gamma-tACS group, we used a fixed 40 Hz as the input frequency. While calculating the IAF, each participant’s individual current intensity was determined through a protocol that began at an intensity of 0.5 mA and was increased in 0.1 mA increments until the participant reported discomfort. This process usually finished within one minute, followed by another one-minute stimulation using the intensity so that participants became habituated to the sensations. The mean intensity was 0.57 ± 0.14 mA (M ± SD).
TACS set-up
Stimulation was administered using an HD-tACS system (Soterix Medical, New York, NY) with a 4 × 1 montage (Fig. 1C and 1D). The central electrode was placed over the F4 electrode, with the neighbouring electrodes over Fz, C4, FT8, and FP2 (International 10–10 Modified Combinatorial Nomenclature). The tACS was delivered at an individual current intensity. Participants were informed about potential altered sensations during stimulation before testing individual current intensity. Furthermore, a one-minute stimulation was administrated to all groups before the 15-minute tACS session to familiarize participants with the sensations elicited by the electrical stimulation. For the IAF tACS and 40 Hz tACS groups, the current was ramped up from zero to each participant’s individual current density, which was then maintained constant for 15 minutes. The sham group received ramped-up stimulation only in the first 30 seconds.
Behavioral Data Analysis
We used R (Version 4.1.3; R Core Team (2020) for statistical analyses and employed linear mixed-effects models (LMMs) to account for various factors and variances. Fixed factors included tACS group, time, and feedback desirability. Consistent with previous research (Sharot et al., 2011; Kuzmanovic et al., 2018; Yao et al., 2021), covariates were trial-wise first estimation (E1), subjective ratings for adverse life events (vividness, personal relevance, familiarity, prior experience), memory error, and differences in trial numbers between desirable and undesirable conditions. Random effects accounted for estimation error (EE), participant and event ID at the trial-level. We used the Satterthwaite’s method (‘anova’ function in the package “lmerTest” ; Kuznetsova et al., 2020) to test significance levels for fixed effects. Post-hoc analyses were performed with the ‘emmeans’ and ‘emtrends’ function in the package “emmeans” (Lenth et al., 2022) to investigate significant interaction effects. Unless otherwise specified, post hoc comparisons were corrected using the false discovery rate (FDR) method.
Trial-level analysis on belief updating
To examine the immediate tACS effect as well as the 24-hour delay effect, we ran the LMM with time (pre, post, delay), desirability (desirable, undesirable), and group (IAF, 40 Hz, sham) as fixed factors to predict belief updating (E2-minus-E1 on Day 1, E3-minus-E1 on Day 2). Consistent with previous research, we added covariates including E1, memory error, trial numbers, and event ratings. Random effects included participant, EE, and event ID (full model definition was presented in SOM).
Participant-level analysis on belief updating
We calculated each participant’s update bias (i.e., averaged desirable updating -minus- averaged undesirable updating for each participant), and conducted a mixed 3 (time, pre, post vs. delay) * 3 (group, IAF vs. 40 Hz vs. sham) ANOVA with the update bias as the dependent measure. Potential confounds were included as covariates, including E1, memory score, trial number difference between desirable and undesirable conditions, and event ratings for vividness, familiarity, prior experience and personal relevance.
In all above analysis, covariates were z-scored within each participant.
EEG Data Analysis
Resting-State EEG Data Preprocessing
Artifact-free epochs from the eyes-open resting-state EEG data were visually inspected and selected. A minimum of 90 s of artifact-free data per phase was required. After selection, ICA was applied directly to the cleaned data, and eye-movement-related components were removed using ICLabel (Pion-Tonachini et al., 2019) in EEGLAB.
Task EEG Data Preprocessing
An unused channel was removed (LZ), leaving data from the remaining 63 channels for the analysis. All EEG processing was performed using MATLAB-2022b and functions from EEGLAB (Delorme & Makeig, 2004) and ERPLAB (Lopez-Calderon & Luck, 2014) toolboxes. The EEG raw data were downsampled to 250 Hz, and notch filtered at 50 Hz using the PMnotch with an order of 180, and a high pass filter of 0.5 Hz was applied using the ERPLAB plugin. The filter order was set to 2. Bad channels were detected using a trim outlier plugin, followed by visual inspection and interpolation. The common average was then calculated by adding a zero channel to make the data full rank. Continuous EEGs were then epoched into [-1000 to 2000 ms] segments relative to the onset of feedback, and to the question mark before the second estimation, without baseline correction. Visual inspection was conducted to remove epochs with intensive artifact-contamination in order to improve performance of the independent components analysis (ICA). To facilitate ICA, we first removed the interpolated channels and filtering the clean epoched data with a 1 Hz high-pass to enhance ICA performance. Subsequently, the ICA weights and sphere were re-applied to the original cleaned epoch data (with interpolated channels removed). We used the ICLabel plugin to remove eye- and muscle-related activity components. The clean ICA-processed data were interpolated for previously removed channels. We then performed an automatic peak-to-peak artifact removal step to handle any remaining epochs with artifacts not captured by ICA. The remaining clean data were used for spectral analysis.
EEG Spectral Analysis
To examine how tACS influenced brain activity during resting and task states, we performed EEG spectral power analysis during eyes-open resting states for a minimum of 90 seconds, and in task states within the first second following the presentation of the feedback probability. We focused on the alpha and gamma frequency bands in the right frontal channels (F2, F4, FC4, F6, AFz). Using Fast Fourier Transforms (FFTs) with Welch’s method and a Hanning window, we analyzed alpha power within the Individual Alpha Frequency (IAF ± 2Hz) and the gamma power (30–50Hz) over these five channels on laplacian current source density (CSD) transformed EEG data.
The Frontal Alpha Asymmetry (FAA) was calculated by obtaining raw power from the left F3 and the right F4 (Smith et al., 2017), which was then log-transformed (ln) to assess the difference in alpha power between these electrodes (ln alpha F4 - ln alpha F3). To evaluate the tACS effect on the right frontal EEG power changes, alpha and gamma power were averaged across the five aforementioned channels (F2, F4, FC4, F6, AFz). For resting states, raw power was utilized over the entire 90-second epoch, while for task states, power was normalized against a baseline period (− 1000 to 0 ms) and expressed in decibel units (10log10) in statistical analysis.
To quantify the specific effects of tACS on EEG activity, we employed Analyses of Covariance (ANCOVAs) focusing on frontal alpha asymmetry (FAA), right frontal alpha, and right frontal gamma power, both during resting and the belief update task (involving feedback encoding and re-evaluation processing phases, Fig. 1B). In these ANCOVAs, we used the pre-tACS values of the corresponding outcome measure as covariates to account for baseline differences, and then compared the post-tACS EEG outcome measures across three groups, respectively. For task EEG analyses, it involved 3 (between-subject, tACS groups) by 2 (feedback desirability) by 2 (processing phase) ANCOVAs for each of the three outcome measures.
To investigate the association between EEG activity changes and belief updating changes from pre- to post-tACS, we ran a series of correlation with changes of FAA/right frontal alpha/gamma power and update changes for desirable and undesirable condition separately. All correlations were FDR-corrected for multiple comparisons for the total number of 12 correlations across three groups separately for FAA/right frontal alpha and gamma power.