A total of 21 healthy graduate students at Fudan University participated in this study (i.e., 11 men and 10 women, age: 26.47 ± 1.91 years). All subjects had no reported history of neurological or psychiatric disorders, had a normal or corrected-to-normal vision, with intact cognitive function as assessed by the total score of mini-mental state examination (MMSE) greater than 27, and were right-handed determined by the Edinburgh Handedness Inventory (Oldfield, 1971). Subjects were asked to avoid any intake of alcohol or caffeine for 24 h prior to testing. Written informed consent form was obtained prior to their participation in the study. The ethical approval of the study was granted by the School of Life Sciences, Fudan University.

Each subject underwent three study visits consisting of the performance of gradCPTs before, during, and after receiving one session of either O-tDCS, C-tDCS, or sham stimulation. Each test lasted for 8 min and was successively conducted before, during the second half of the 20-min stimulation, and after stimulation. Each session lasted for about 40 min (Figure 1A). A questionnaire about tDCS side effects was filled in at the end of each session. Of note, these successive tests might trigger the practice effect or fatigue effect, reducing some experimental effects, e.g., vigilance decrement. Therefore, three stimulation conditions were conducted in a randomized order to balance the possible impact of practice effect or fatigue effect on stimulation conditions. Furthermore, these three visits were separated by at least 72 h between each to eliminate the potential carryover effect of the prior stimulation. The participants were blinded to the type of tDCS. The C-tDCS and sham served as the active control condition and baseline condition, respectively.

The gradCPT represents a unique combination of task features, in that, it requires frequent overt responses and removes abrupt stimulus onsets that may exogenously capture attention (Rosenberg et al., 2013). It was selected to measure sustained attention due to its very high reliability for both behavioral and neural measurements compared with other sustained attention tasks (Rosenberg et al., 2016). As shown in Figure 1B, stimuli in the gradCPT were round, grayscale photographs containing 10 mountain scenes and 10 city scenes. These scenes were randomly presented with 90% city (i.e., the go stimulation) and 10% mountain (i.e., the no-go stimulation), without allowing the identical scene to repeat on consecutive trials. The gradual transition of one image to another used the linear pixel-by-pixel interpolation, where the complete transition occurred over 800 ms. Participants were asked to press the space bar on the keyboard for each city scene but withhold their responses for mountain scenes. The reaction time (RT) was defined as the delayed time of response relative to the beginning (0%) of each image transition. The response accuracy was emphasized without reference to speed. However, given that the next image would replace the current image in 800 ms, a response deadline was implicit in the task (refer to RT analysis). The MATLAB software (MathWorks) and Psychophysics Toolbox (Brainard, 1997) were used to present stimuli and collect responses.

The tDCS was delivered with the 128-channel Geodesic Transcranial Electrical Neuromodulation (GTEN) system (Philips Neuro, Eugene, OR, United States). For the C-tDCS, the 1 mA total constant current was delivered for 20 min through five prefrontal source channels (the maximum current density was focused on the left DLPFC) to five sink channels, located at the parietal-occipital area (Figure 1C). Source/sink electrodes were positioned at the same location across three stimulation conditions. The time for ramping up and ramping down was 15 s at the beginning and end of the stimulation, respectively. For the O-tDCS, the stimulation current fluctuated between 0 and 1 mA, with 5 s on/5 s off periods, and rising and falling slopes of 5 s; thus, resulting in a 0.05 Hz oscillating stimulation, totally applied for 20 min. For the sham condition, no constant current was delivered except at the first and last 15 s, respectively (Figure 1D). To minimize sensation from the current injection, the Elefix conductive paste for the stimulating electrodes was mixed with a lidocaine solution for both stimulation and sham.

The RTs were analyzed following previous studies about gradCPT (Esterman et al., 2013, 2014b). Since the RT was calculated relative to the beginning of each image transition, an RT of 800 ms indicates a button press at the moment an image was 100% coherent and not mixed with the successive image. An RT shorter than 800 ms indicates that the current image was still in the process of transitioning from the previous, whereas an RT longer than 800 ms indicates that the current image was in the process of transitioning to the subsequent image. On rare trials with highly deviant RTs (before the 70% coherence of image n or after the 40% coherence of image n + 1) or multiple button presses, an iterative algorithm maximized correct responses as follows. First, the algorithm assigned unambiguous correct responses, leaving few ambiguous button presses (<5% of trials). Second, an ambiguous press was assigned to an adjacent trial if one of two successive trials had no response. If two successive trials had no response, the press was assigned to the closest trial, unless one was a no-go target, in which, case subjects were given the benefit of the doubt that they correctly omitted. Third, if there were multiple presses that could be assigned to any one trial, the fastest response was considered a valid response. Finally, if more than two successive trials had no response and some trials cannot be assigned with the proper response, the missing RTs to cities and mountains were filled up with the median of RTs of each type of scene (city or mountain) in each run, respectively. After those processes, the mean and standard deviation (SD) of RTs for each run were calculated. The lower mean and SD of RT represented higher attention focus and attention stability, respectively (Yamashita et al., 2021).

Trials in which participants responded to mountains were considered commission errors. The failure of response suppression to mountains reflects a lower inhibitory control level. Trials in which participants failed to respond to cities were considered omission errors, which is possibly due to the lack of attention focus or attention lapses (Egeland and Kovalik-Gran, 2010).

According to previous experience in parameter optimization (Esterman et al., 2013), vigilance decrements were calculated with a 2 min sliding window around performance measures of interest (i.e., commission error, omission error, RT_mean, and RT_SD), where the first window contained 0–2 min and the last contained 6–8 min. The window moved with a step of 1 trial. A linear slope (computed as the rate of change per minute) was then calculated for each run. Vigilance decrements were determined if slopes are larger than zero in one-sample t-tests.

The time series of RTs in each run was transformed to the frequency domain with the fast Fourier transform (FFT) (Wang et al., 2016). The frequency resolution was 0.0021 Hz (sampling rate/sampled data: 1.25 Hz/600). The power spectrum of each run was obtained to test whether the O-tDCS evoked low-frequency behavioral oscillations.

To evaluate whether the enhanced sustained attention was driven by the enhanced power of behavioral oscillations, we used the value of power at 0.05 Hz to predict other indicators of sustained attention. The leave-one-out approach was used to estimate the predicted value of each indicator of each subject. Indicators under the O-tDCS and sham conditions during stimulation were recruited in the linear regression model to construct the relationship between power at 0.05 Hz and the commission error, omission error, RT_mean, and RT_SD. The least-square method was used to determine the predicted values of commission error, omission error, RT_mean, and RT_SD.

The three stimulation types (i.e., O-tDCS, C-tDCS, and sham) by three tests (i.e., pre-stim, in-stim, and post-stim) repeated measures analysis of variance (ANOVA) was performed on the power of each frequency point. The 3×3 ANOVA was also performed on remaining indicators, including the mean and SD of RTs, commission errors, omission errors, and slope. The post hoc analysis was conducted with the paired-sample t-test if there was a significant interaction between the stimulation type and test. Since predicted values were lower than observed values, Spearman’s correlation was performed to test the relationship between observed values and predicted values of the commission error, omission error, RT_mean, and RT_SD (Rosenberg et al., 2016). The Greenhouse-Geisser correction was used if the equal variance was not assumed. Bonferroni’s correction (p < 0.05) was used if there were multiple comparisons for each indicator.

As shown in Figure 2A and Table 1, significant main effects of stimulation and test as well as their interaction for commission errors were observed, suggesting that stimulations enhance inhibitory control. The post hoc analysis revealed significantly reduced commission errors by the O-tDCS than the C-tDCS [t(20) = −2.727, p = 0.013, Cohen’s d = −1.22 for in-stim, t(20) = −5.62, p < 0.001, Cohen’s d = −2.513 for post-stim] and sham [t(20) = −5.555, p < 0.001, Cohen’s d = −2.484 for in-stim, t(20) = −5.274, p < 0.001, Cohen’s d = − 2.359 for post-stim] as well as marginally significant reduction of commission errors by the C-tDCS than sham [t(20) = −2.159, p = 0.043, Cohen’s d = −0.966 for in-stim, t(20) = −2.82, p = 0.011, Cohen’s d = −1.261 for post-stim]. Compared with pre-stim, the commission errors were reduced only by the O-tDCS under in-stim and post-stim [pre-stim vs. in-stim: t(20) = 6.173, p < 0.001, Cohen’s d = 2.761; pre-stim vs. post-stim: t(20) = 5.724, p < 0.001, Cohen’s d = 2.56].

Similarly, the main effect of stimulation and the interaction of stimulation by test for omission errors were significant (Figure 2A and Table 1), indicating that stimulations reduce attention lapses. The omission errors were reduced only by the O-tDCS compared with sham [t(20) = −2.709, p = 0.014, Cohen’s d = −1.212 for in-stim, t(20) = −4.68, p < 0.001, Cohen’s d = −2.093 for post-stim]. Compared with pre-stim, the omission errors were reduced only by the O-tDCS under in-stim [t(20) = −2.439, p = 0.024, Cohen’s d = −1.091] and post-stim [t(20) = −3.066, p = 0.006, Cohen’s d = −1.371].

Overall, the O-tDCS reduced both commission errors and omission errors than the C-tDCS and sham. The modulation effect of O-tDCS on commission errors and omission errors was even larger after stimulation than during stimulation. The commission errors and omission errors were lower under C-tDCS than under sham, but no significant difference was achieved.

Although there was only a significant main effect of stimulation for the SD of RT, the interaction of stimulation and test was significant for both mean and SD of RT (Table 1 and Figure 2B), suggesting that stimulations improve attention focus and stability. The post hoc analysis showed that only the O-tDCS significantly reduced both the mean [t(20) = −2.939, p = 0.008, Cohen’s d = −1.314 for in-stim, t(20) = −2.672, p = 0.015, Cohen’s d = −1.195 for post-stim] and SD [t(20) = −4.552, p < 0.001, Cohen’s d = −2.036 for in-stim, t(20) = −3.296, p = 0.004, Cohen’s d = −1.474 for post-stim] of RT, while the C-tDCS reduced the SD [t(20) = −2.222, p = 0.038, Cohen’s d = −0.994 for in-stim] of RT with marginal significance compared with that in sham. In contrast, only the O-tDCS reduced the mean and SD of RT under in-stim [t(20) = −3.75, p = 0.001, Cohen’s d = −1.677 for mean, t(20) = −3.334, p = 0.003, Cohen’s d = −1.491 for SD] and post-stim [t(20) = −3.142, p = 0.005, Cohen’s d = −1.405 for mean, t(20) = −2.894, p = 0.009, Cohen’s d = −1.294 for SD] compared with pre-stim.

There was no stimulation effect for the slope as all slope values were close to zero (mean = -0.002; ranged from −0.177 to 0.104). In other words, there was no vigilance decrement in this study. However, the sliding window analysis replicated the aforementioned results about attention lapses, attention focus, attention stability, and inhibitory control that the O-tDCS reduced all accuracy and RT indicators under in-stim and post-stim (Figure 3), suggesting that the O-tDCS enhances sustained attention across the time of the whole task.

The O-tDCS evoked remarkable oscillations in the RT around the stimulation frequency. Significant interaction of stimulation and task was observed within 0.031–0.063 Hz, peaking at 0.05 Hz (Figure 4A). During stimulation, the power of RT oscillations at the stimulation frequency was tremendously enhanced by the O-tDCS more than 4 times of the C-tDCS [t(20) = 76.45, p < 0.001, Cohen’s d = 34.189] and sham [t(20) = 90.897, p < 0.001, Cohen’s d = 40.65], which reduced to about 3 times of the C-tDCS [t(20) = 46.11, p < 0.001, Cohen’s d = 20.621] and sham [t(20) = 48.676, p < 0.001, Cohen’s d = 21.769] after stimulation. Only the O-tDCS enhanced the power of RT oscillations at 0.05 Hz for in-stim [t(20) = 70.699, p < 0.001, Cohen’s d = 31.618] and post-stim [t(20) = 48.393, p < 0.001, Cohen’s d = 21.642] compared with pre-stim (Figure 4B).

Further analysis revealed that the power at 0.05 Hz effectively predicted commission error, omission error, RT_mean, and RT_SD during stimulation. The correlations between observed values and predicted values were all significant [r = 0.601, p < 0.001 for commission error; r = 0.516, p < 0.001 for omission error; r = 0.37, p = 0.016 for RT_mean, and r = 0.509, p = 0.001 for RT_SD].