Tailoring transcranial alternating current stimulation based on endogenous event-related P3 to modulate premature responses: a feasibility study

Participants

Twelve healthy volunteers (seven females; mean age: 25.67 ± 1.97) participated in the study and signed the informed consent before their enrollment in the study. The participants were right-handed and without any history of neurological and psychiatric disorders. The study was approved by the University of Minho ethics committee (Ethics Committee for Research in Life and Health Sciences; CEICVS 057/2021) and we followed the recommendations of the Declaration of Helsinki regarding ethical standards and participants’ safety.

Study design

The study was performed in two sessions with distinct tACS conditions, namely active and sham stimulation. The order of the stimulations was counterbalanced and randomly assigned to the participants. Sessions were scheduled at least 48 h apart to avoid carryover effects. In the first session, self-report questionnaires were administered to evaluate handedness, impulsive traits, and clinical symptomatology (see Table S1). Additionally, in the first session, participants performed a baseline block of the Cued Premature Response Task (CPRT) concomitant with the EEG data collection, to assess the latency and the ERO associated with the P3 component. Afterwards, three blocks of the waiting impulsivity task were performed, in which the first and the third ones were performed only with EEG (i.e., pre and post-tACS), while the second block was performed concomitantly with tACS (see Fig. 1A). The second session did not include the baseline block; thus, participants only performed the last three experimental blocks (i.e., pre, during, and post-tACS). At the end of each session, participants completed a blinding questionnaire (see results in Table S2).

Cued premature response task

The CPRT was programmed and executed in E-Prime 3 software (Psychology Software Tools, Pittsburgh, PA, USA) based on a pre-print by Mendes et al. (2024). The baseline block performed only in the first session comprised 10 training trials and 100 experimental trials. Moreover, the following three experimental blocks also comprised 100 experimental trials each. The total duration of each block was approximately 10 min, totaling 40 min in the first session (i.e., 30 min in the second session), including several pauses between the experimental blocks. Participants were asked to press a button in the E-Prime Chronos response box to begin the experiment, and to release it as soon as possible following the target appearance on screen. The target was always preceded by a cue, which informed the participant that the target was about to be displayed. In the first and last blocks (i.e., pre and post-tACS), the interval of time between the start of the trial and the cue was jittered between 1000 and 1500 ms, while the interval between the cue and target was jittered between 500 and 2500 ms (see Fig. 1B). On the other hand, the block of CPRT during tACS had a “wait” adjustment that was dependent on the stimulation parameters that were individually estimated in the EEG online analysis of the baseline block (see ‘Electrophysiological acquisition and data analysis’) (Dallmer-Zerbe et al., 2020). This is of particular interest since the tACS needs to match the temporal properties of the oscillatory activity. We aimed to entrain electrophysiological activity that occurred in a specific period after target onset (Jones, 2016). In this block, the interval between cue and target was randomly selected between 300 and 2300 ms plus the “waiting” period (see Fig. 1C), to ensure the synchronization between the peak of tACS and the target P3 latency, thus minimizing “drifts” between the onset of stimulation and the stimulus presentation. This synchronization was controlled in MATLAB (MathWorks, Natick, MA, USA) with the MatNIC package to trigger the stimulation by the Starstim R20 (Neuroelectrics, Barcelona, Spain). MATLAB initiated stimulation by transmitting a UDP trigger to the Python console, which subsequently transmitted iterative TCP triggers to E-Prime 3 whenever the tACS reached phase 0 (please check the MATLAB and Python scripts with the E-prime file of CPRT in the Open Science Framework (OSF) repository). Please refer to Figs. S1 and S2 for some examples.

The reinforcement/punishment feedback pretended to elicit a higher number of premature responses. Therefore, the participant’s response was rewarded with virtual money if his/her response was fast, punished if his/her response was slow, or neither rewarded nor punished if they released the button before the target onset (i.e., premature response). The feedback was tailored for each participant based on the mean and variability of the response time (RT) observed in the last 10 trials of the baseline block. The mean and variability of the response time (RT) observed in the last 10 trials of the baseline block were considered to estimate the reinforcement/punishment feedback (see Fig. SM3), specifically:

• Very fast responses: if the participant released the button with an RT below −0.66 standard deviation (SD) of the baseline RT mean, the participant was reinforced with virtual 1€. Moreover, if any participant earned 1€ three times in a row, the feedback increased to 2€ for “very fast responses”.

• Fast responses: if the participant released the button with an RT between −0.66 SD and +0.33 SD of the baseline RT mean, the participant earned a virtual 0.5€.

• Slow responses: if the participant released the button with an RT between +0.33 SD and +1SD of the baseline RT mean, the participant lost virtual 0.5€.

• Very slow responses: if the participant released the button with an RT above +1SD of the baseline RT mean, the participant lost virtual 1€.

• Premature responses: if a participant released the Chronos button before the target, the feedback was “Continue”, in a way that participants were not reinforced nor punished.

Transcranial alternating current stimulation

tACS conditions were applied in distinct sessions through the Starstim R20 (Neuroelectrics, Barcelona, Spain). The 25 cm² round saline-soaked electrode sponges (∼radius of three cm, current density: 0.08 mA/cm2) were placed in two clusters of two electrodes each, specifically in the parietal areas (i.e., P3 and P4) and supraorbital area (i.e., Fp1 and Fp2) (see Fig. 1D). The clusters delivered alternatingly anodal and cathodal stimulation to ensure sinusoidal stimulation (Popp et al., 2019). The active tACS was delivered at 2 mA (peak-to-peak) intensity for 10 min (with 15 s of ramp up and ramp down) during the CPR Task. tACS frequency was individualized for each participant in the ranges of delta and theta band (1.5–7 Hz) according to the peak ERO power detected in the baseline block (check Table S3 for individual information). The average stimulation frequency was 3.29 Hz (SD = 1.9 Hz) in line with previous studies, using a similar methodology (Dallmer-Zerbe et al., 2020). For the sham procedure, tACS was delivered for 15 s (with 15 s of ramp up and ramp down: 45 s in total) at the beginning and the end of the 10-minute session.

Electrophysiological acquisition and data analysis

EEG data collection was performed with the Starstim R20 (Neuroelectrics, Barcelona, Spain) in the following electrodes: F7, F3, Fz, F4, F8, C3, Cz, C4, T7, P7, Pz, T8, P8, O1, Oz, and O2. Data was sampled at a rate of 500 Hz and analyzed using the EEGLAB toolbox (Delorme & Makeig, 2004a) at two specific times: online and offline analysis. The online analysis was performed between the baseline block and the first experimental block of the first session, whilst the offline comprehended the remaining EEG data, namely pre and post-tACS blocks from both sessions.

The EEG analysis started with filtering between 0.5 and 40 Hz using an FIR filter and the line noise (i.e., 50 Hz) using a notch filter before re-referencing to the average. This continuous data was epoched around the cue and the target with a total duration of four seconds (i.e., 2000 ms before and after the stimulus). A baseline correction considering the 200 ms before the target onset was performed. The cue epochs with a cue-target interval lower than 800 ms or with a premature response in the initial 1000 ms were removed from the analysis. Moreover, epochs exceeding ±100 µV in the Cz or Pz during the time window of interest (i.e., between −200 ms and 600 ms) were also removed. Finally, a visual inspection was also performed to detect potential artifacts not been removed before. The rejection rate was approximately 10% of the epochs, which followed the recommendations by Delorme, Sejnowski & Makeig (2007) (see Table S4).

The offline data analysis was similar to the online plus the artifact removal through the ICA. This step was only performed in the offline analysis considering the required computational time and comprehensive examination of each component. The grand averaged ERPs in the figures were filtered using a 12 Hz low-pass filter. One participant was removed from the data analysis because the post-tACS file of the sham session was corrupted.

Sample size calculation

The sample size calculation was based on an effect estimate of 1.2 (Cohen’s d) between active and sham sessions observed in the preliminary study testing tACS in P3 amplitude (Dallmer-Zerbe et al., 2020). Thus, the sample size was estimated to find a within-group effect between active and sham tACS in P3 amplitude with a statistical power of 95% and an alpha level of 5%. A total of 10 participants were calculated, but we added two participants to the estimation considering potential differences with the aforementioned study. The final sample size for this feasibility study was 12 participants. The calculation was performed in G*Power version 3.1.9.7 (Faul et al., 2009).

Statistical analysis

The absolute change was calculated between pre and post-tACS for each session (i.e., active and sham). For that, the values observed in the block before tACS were subtracted from the values obtained after tACS. This estimation was performed in every EEG and behavioral outcome. We followed a similar procedure to the one followed by Dallmer-Zerbe et al. (2020). However, we chose the absolute change instead, because relative change may lead to increased variability when there are very low values in the pre-tACS block. Paired t-tests were used to assess the absolute change observed between active and sham sessions if the difference between both tACS conditions followed a normal distribution according to the Shapiro–Wilk test. Otherwise, non-parametric analysis was performed, specifically the Wilcoxon signed-rank test. A p-value of 0.05 in a two-tailed test (0.025 for each direction) was considered the significance threshold for all analyses. The statistics were analyzed in R (R Core Team, 2020).

Event-related potentials: P3

The nonparametric analysis revealed a significant difference in the absolute change between active and sham sessions in target-P3 amplitude (W(10) = 63, p = 0.01), but no significant effect was detected for the cue-P3 (t(10) = −0.11, p = 0.91). In the additional analysis considering the temporal adjustment around P3 latency to each participant (i), there was a significant effect in target-P3 amplitude between active and sham tACS (t(10) = 2.84, p = 0.02). The absolute change of target-P3 amplitude was higher in the active session in comparison with the sham in both analyses (Fig. 2B; Table 1).

Event-related oscillations

No significant effects between both sessions were observed in evoked-delta (t(10) = 1.55, p = 0.15) and evoked-theta (t(10) = 0.26, p = 0.79) during target-P3 (Fig. 2C; Table 1). On the other hand, paired t-tests revealed a marginal significant effect in delta activity during cue-P3 (t(10) = −2.06, p = 0.07), but not for theta (t(10) = −1.54, p = 0.15) (Fig. 3C; Table 2). Furthermore, the additional analysis of the temporal adjustment (i) did not reveal significant differences in delta (t(10) = 1.59, p = 0.14) and theta (t(10) = 0.79, p = 0.45) during target-P3 between sessions. In the frequency adjustment analysis (ii), no significant differences were detected in the adjusted ERO from target-P3 (t(10) = 0.98, p = 0.35), but, a significant effect between both tACS session in the adjusted ERO from cue-P3 (t(10) = −2.07, p = 0.03) was found. The active tACS session showed a significant decrease in the adjusted ERO in comparison with the sham. Finally, the temporal and frequency analysis performed in target-P3 (iii) did not reveal a significant effect on adjusted ERO between active and sham (t(10) = 1.11, p = 0.29).

Power spectral analysis

No differences were found in terms of the absolute change between active and sham sessions for the delta (W(10) = 37, p = 0.77) and theta bands (W(10) = 45, p = 0.32) in target-P3 epochs (Fig. 2D; Table 1). Likewise, no significant differences in spectral power of cue-P3 epochs for delta (t(10) = 0.04, p = 0.97) and theta (t(10) = 0.46, p = 0.66) (Fig. 3D; Table 2) were found. Finally, regarding the frequency adjustment analysis (ii), there were no significant differences between both sessions for the target-P3 (t(10) = 0.73, p = 0.48) and the cue-P3 epochs (t(10) = 0.61, p = 0.56).

Behavioral analysis

The paired t-test did not reveal statistically significant effects in the number of premature responses between the active and the sham session (t(11) = −0.51, p = 0.62). Likewise, the nonparametric analysis also did not show significant effects in the absolute change of total money earned/loss (W(11) = 41, p = 0.51) and release time (W(11) = 41, p = 0.91) between both sessions (Table 3).

Limitations and future directions

The sample size calculation was aimed at detecting an effect in P3 amplitude, which resulted in a low sample size to test other hypotheses (e.g., decrease of premature responses). In addition, our study did not replicate exactly all the same methods employed in previous studies. Namely, we performed a within-subjects design and used absolute change instead of relative change. However, the descriptive data suggests an increase in ERO during the active tACS session and a decrease in sham (see Table 1; Fig. 2), as similarly observed by Dallmer-Zerbe et al. (2020). Therefore, the effects on ERO activity should be further tested in studies with larger sample sizes, including both hypotheses raised for the ERO decrease in cue-P3. In specific, future studies should address how the temporal and frequency mismatch between stimulation and EEG endogenous activity (i.e., anti-phasic and STDP hypothesis respectively) can decrease ERO power and consequently reduce ERP amplitude.

Moreover, this study tested the offline effects of tACS through the absolute change between the blocks performed before and after the stimulation, rather than analyzing the block during the stimulation (i.e., online effects). The literature on tACS suggests offline aftereffects with a duration of at least 30 min (Neuling, Rach & Herrmann, 2013) or 70 min (Kasten, Dowsett & Herrmann, 2016). On the other hand, other studies have suggested that tACS effects are commonly observed online, rather than offline (Pozdniakov et al., 2021). Likewise, P3 latency might change over time (Devos et al., 2020), which can represent an obstacle to correct temporal synchronization because we only evaluated endogenous EEG activity for each subject in the first session. However, although we did not observe significant fluctuation of P3 latency during the experimental session (see Table S5), future research should focus on closed-loops stimulations that are dependent on the online endogenous oscillatory activity to optimize tACS effects (Frohlich & Townsend, 2021; Leite et al., 2017). This can also ultimately allow to increase the precision of tACS on the behavioral outcomes, as the stimulation could also be adjusted based on individual performance. For instance, due to our sample size, we were unable to find differences in terms of release times; however, it seems that active tACS increased it, while sham tACS decreased it. It also seems that while there was an increase in the variability of the responses in the active condition, there was a decrease (when compared to baseline) for the sham condition. Therefore, future studies should take into account these performance-based closed loops, to increase the precision of the effects of tACS in cognitive, as well as behavioral outcomes.