Forty-nine healthy participants were recruited in this study (15 males and 34 females; aged 20 to 30 years, mean age = 21.96 years). Before being recruited into this study, a safety screening was performed to identify and exclude any participants with contraindications to tRNS. None of the participants had neurological or psychiatric issues and none had taken psychoactive drugs in recent months. Participants were randomly assigned to the tRNS group (n = 24) or sham group (n = 25). All experimental procedures were approved by the Joint Institutional Review Board of Taipei Medical University, and all participants gave informed consent prior to their participation.

The experiment took approximately 2 hours to complete. Resting-state EEG data were collected before and after the stimulation on the same day. Participants were told to relax while EEG recording was in progress. EEG data were collected with eyes open for 3 minutes (fixating on a fixation cross in front) and closed for 3 minutes, for 3 cycles (18 minutes total). Participants then received 20 minutes of full spectrum tRNS or sham stimulation. Following that, another 18-minute post-stimulation resting-state EEG session, identical to the pre-stimulation session, was performed.

Electrical stimulation was applied using a StarStim stimulator (Neuroelectric, Barcelona, Spain). Two 25 cm${}^2$, circular shaped rubber electrodes covered with saline (NaCl solution)-soaked sponges were placed in the middle of Fp1/Fp2 and Oz. These fronto-occipital montages were chosen to spread out the electrodes, thereby enhancing the visibility and detectability of tRNS-related changes in EEG signals in any particular region (as shown in Fig. 1). Stimulation with tRNS was applied for 20 minutes with random noise stimulation (0 to 500 Hz) using a current of 1-mA peak-to-peak (–0.5 mA to 0.5mA), which has been shown to produce maximal effect in electric stimulation [45, 46]. The electrodes had no offset, and the stimulation started and ended with a 30-second on- and off-ramp. The sham condition consisted of the same on- and off-ramp, but no actual stimulation was conducted.

Fig. 1.

Montage and model of current flow. Electrodes were placed over the middle of Fp1/Fp2, and Oz. Modeling of current flow was produced using StarStim.

The EEG signals were continuously recorded from 32 Ag/AgCl electrodes (10/20 system) with a reference electrode between Fz and Cz. All electrodes were mounted on a BrainCap electrode cap (Brain Products GmbH, Munich, Germany). All signals were amplified using a BrainAmp amplifier (Brain Products GmbH, Munich, Germany) and recorded using a BrainVision Recorder (version 1.22, Brain Products GmbH, Munich, Germany). The signals were digitized at a sampling rate of 500 Hz. Two sets of electrodes were placed on the upper and lower sides of the right eye, and the canthi of both eyes, to measure participants’ vertical and horizontal eye movements. Electrode impedance was maintained below 10 k$\mathrm{\Omega }$.

All preprocessing and analysis procedures were conducted using EEGLAB version 2021.1 [47] and custom MATLAB (R2020b; MathWorks Inc., Portola Valley, CA, USA) scripts. Our preprocessing showed that data from the eyes-closed blocks were too noisy, with ocular artifacts due to participants’ excessive eye movements, and thus the eyes-closed blocks were excluded from further analysis. Eyes-open EEG data were merged. An independent component analysis was performed to remove ocular artifacts from EEG signals. The continuous ocular-corrected EEG data were first offline and re-referenced to the average of the electrodes at the left and right mastoids (A1 and A2). EEG data were then segmented into 2 second epochs. Epochs with artifacts fluctuating over $\pm$ 100 µV on the channels were rejected. Power spectra of artifact-free epochs were computed using the EEGLAB function “Spectopo”. The spectral power values for the frequencies were then averaged across the canonical EEG bands: theta (4–7 Hz), alpha (8–12 Hz), beta (13–30 Hz), and gamma (31–55 Hz).

A mixed analysis of variance (ANOVA) was conducted to test if power changes between pre- and post-stimulation occurred between the sham and tRNS groups in each frequency band. There was one within-subjects factor of time and one between-subjects factor of the stimulation group. Four scalp regions were chosen to perform the statistical analysis, creating the within-subjects factor of anterior-posterior electrodes: frontal (F3, Fz, and F4), central (C3, Cz, and C4), posterior (P3, Pz, and P4), and occipital regions (O1, Oz, O2). The Geisser–Greenhouse correction was applied when the assumption of sphericity was violated. Holm correction was applied for post-hoc tests. Findings from the ANOVA can be found in Supplementary Results.

In addition to spectral power, we measured phase synchronization between the two stimulation sites, Fz and Oz. The inter-site phase coherence is defined by

$$\text{Intersite Phase coherence}=\left|n^{-1}\sum _{t=1}^n{e}^{i\left({\mathrm{\varnothing }}_{xt}-{\mathrm{\varnothing }}_{yt}\right)}\right|$$

Where n represents the number of time points, and ${\mathrm{\varnothing }}_x$ and ${\mathrm{\varnothing }}_y$ represents the phase angles from electrodes x and y at frequency f. We averaged phase angle differences between Fz and Oz over time. The inter-site phase coherence ranges between 0 and 1, where 0 signifies the absence of synchronization among two channels, and 1 indicates perfect synchronization. These data were input to a two-way ANOVA to analyze the time-averaged phase coherence between Fz and Oz, considering the factors of time (pre vs post) and group (tRNS vs sham) across four frequency bands, respectively.

Previous studies including those from our own lab have shown variations in individual receptivity to brain stimulation [7, 10, 49, 50]. Specifically, we have repeatedly observed that low-performers (defined by the cognitive task of interest from the sham session) tend to benefit from tDCS and tACS on their cognitive performance, whereas high-performers do not. In the absence of a cognitive task, this grouping procedure was not possible in the present study. However, to examine whether there is any indication of variations in individual receptivity to tRNS, we approximated each participant’s responsiveness to tRNS by performing a “post – pre” contrast for all individual participants across all frequencies. A positive difference would somewhat crudely indicate a positive responder to tRNS. To this end, positive contrasts were observed in 16 out of 24 participants in overall power, 15 out of 24 participants in gamma band power, 16 out of 24 participants in beta band power, 14 out of 24 participants in alpha band power, and 15 out of 24 participants in theta band power. Only 6 out of 24 participants showed full-range positive responses (i.e., $>$0 in all frequency bands mentioned above), whereas 2 out of 24 participants showed all negative responses, and the remaining showed inconsistent responses. A similar pattern was also observed in the sham group, where positive contrasts were observed in 16 out of 25 participants in overall power, 10 out of 25 participants in gamma band power, 13 out of 25 participants in beta band power, 17 out of 25 participants in alpha band power, and 14 out of 25 participants in theta band power. None of these numbers are significantly different from the responder headcounts from the active tRNS group (overall: X${}^2$(1, 49) = 0.038, p = 0.845; gamma: X${}^2$(1, 49) = 2.481, p = 0.115; beta: X${}^2$(1, 49) = 1.09, p = 0.296; alpha: X${}^2$(1, 49) = 0.492, p = 0.483; theta: X${}^2$(1, 49) = 0.214, p = 0.644). As such, although the evidence for robust individual differences in tRNS receptivity is not strong, there were indeed a few individuals who showed a full-range increase in EEG power.

If tRNS is highly interactive with cognitive tasks, one would reason that similar stimulation protocols coupled with different tasks may yield quite different results. Although there are no two studies with 100% identical parameters and task design, a review of some similar studies does support the idea of task-dependent, or task-interactive, tRNS outcome. Taking studies applying tRNS over the temporal cortex as an example, Rufener et al. [37] found that, when a near-threshold sound stimulus was used, temporal processing (using a gap detection task) could be improved with a reduced peak latency of P50 and N1, which approximates to the alpha-beta range in time frequency. In contrast, using similar stimulation parameters, Van Doren et al. [36] found increased power in 40 Hz ASSR instead. This discrepancy is likely attributable to the difference in tasks (auditory gap detection task vs ASSR) as the stimulation protocol is highly similar across both studies.

Under this framework, the interactive nature of tRNS can also be observed where tRNS reverses the oscillatory effect behind certain cognitive functions. For example, previous studies have revealed a negative correlation between task performance (e.g., attentional control and response inhibition) and TBR [54, 55]. Behzadnia, Ghoshuni and Chermahini [56] showed an increase in TBR when participants performed a visual version of the conjunctive continuous performance task compared with eyes-open resting EEG. However, one recent study applying frontal tRNS coupled with a continuous-monitoring task actually showed reduced TBR compared with sham stimulation [39]. tRNS-induced neuromodulation therefore likely depends on the task at hand and thus yields different (or opposite) results across studies.

Another potential interactant of tRNS outcome is individual receptivity. For example, Van Doren et al. [36] demonstrated 40 Hz ASSR increase after high-frequency tRNS over the temporal cortex. However, Schoisswohl et al. [44] could not replicate the same findings using both high and low-frequency tRNS. This suggests that tRNS may produce various effects between individuals. This idea is exemplified by a recent study by Rufener et al. [35], who gave participants either tRNS or white noise while participants performed an auditory detection threshold task. These authors observed considerable individual variation in their receptivity to tRNS, and the inverted U-shape function that is predicted by stochastic resonance account is only observed in some, but not all, participants. Our exploratory analyses on individual variations in different frequencies also support their conclusion.

Like tACS, one factor to consider is participant innate baseline activity (or cognitive performance). Harty and Cohen Kadosh [39] recently showed that people with higher TBR baseline could gain more benefit from tRNS over the right frontal and parietal cortex on a sustained attention task. They suggested that this might be due to the number of (relatively) more task-relevant subthreshold neurons that are excited by tRNS [39]. In contrast, people with high mathematic proficiency at baseline do not gain benefits from tRNS and may even show impaired performance in calculation tasks [57]. As such, it seems that tRNS-induced improvement may only be available to those whose neural or cognitive baseline falls within a specific range, and future studies should employ pre-stimulation baseline EEG or cognitive scores to investigate this possibility.

Stochastic resonance is a phenomenon that weak signals can be detected by adding random noise. In vitro studies also support the noise-enhanced mechanism by revealing that more action potentials are generated by adding appropriate noises in the dorsal root ganglion and pyramidal neurons [28, 29]. Applying adequate levels of noise by tRNS can optimize the signal-to-noise ratio in our neural system [58]. However, our results did not show such aftereffects of neural enhancement on EEG, suggesting that a task is important to induce the modulation of brain activity. This makes sense in our study because tRNS noise without any signal (i.e., task-induced activities) would end up being unhelpful noise. Our current observations from EEG are therefore consistent with the predictions of stochastic resonance account.

A study by Shalev et al. [59] showed that attentional selectivity can be boosted by combining bi-parietal tRNS with a spatial priming task. Interestingly, in their study, tRNS was no longer helpful when the spatial priming task was eliminated. Moreover, Cappelletti et al. [60] showed that the effect of tRNS is much smaller when parietal tRNS is applied alone without the aid of cognitive training. Taken together, we can infer that tRNS-induced modulation perhaps needs a specific task to maximize its effects. However, it is important to note that many recent studies have begun to question the validity of stochastic resonance in the context of tRNS and sensory/cognitive processing [35]. Although our null result is well predicted by stochastic resonance, more research is therefore needed to test how well stochastic resonance predicts other task-related tRNS effects.

Although the present study failed to find any tRNS-induced changes in resting state EEG, it remains possible that our study lacked the sample size or statistical power needed to detect the tRNS-EEG effect, if any. Caution is therefore required when considering null results as evidence for the absence of an effect. Nonetheless, our sample size (i.e., 24 in the tRNS condition, 49 in total) is on par with the sample sizes of most task-related tRNS studies, which have achieved significant cognitive-tRNS effects with sample sizes of 8 [30], 10 [25, 60], 14 [36], 15 [23, 24], 16 [17, 19, 41, 61], 17 [16], 18 [18], 19 [34], 20 [37, 38], and 22 [22, 44]. As such, we believe that if there really is an effect that we could not detect with our sample size, such effect (if any) is likely to have a much smaller effect size than those observed from previous task-related EEG studies. This would be consistent with our current conclusion that tRNS effect is more robust when a sensory/cognitive task is present.

Alternatively, it remains possible that other parameters (e.g., montage, current intensity) may be more successful in finding a tRNS-induced resting EEG effect. Although in this study we did try to match current intensity and ramp up/down time as did most previous studies, our choice of fronto-occipital montages (for the purpose of visualizing the current flow if there was an effect) is somewhat unconventional. To our knowledge, only one study has used this setup; Battaglini et al. [62] used fronto-occipital tRNS and observed increase arousal in a discriminative reaction time task (also see [63]). Thus, the effect was also task based in a fronto-occipital design. As such, it remains possible that different results might be obtained with another set of montages or parameters.