The sample size for this study was estimated based on a previous study that utilized HD-tDCS on chronic stroke participants with the same outcome variable (i.e., CMC) (Bao et al., 2019). The value of the effect size ( η p 2 = 0.191 ) was observed in the previous study, which is equal to the effect size value of f (f = 0.56). Considering the differences between healthy and stroke participants (i.e., the modulatory effect of tDCS may be weaker in healthy participants) (Hendy and Kidgell, 2013), there is a conservative reduction in effect size in the current study. Thus, the sample size was calculated using G*Power 3.1.9.7 software (Franz Faul, University of Kiel, Kiel, Germany) with a statistical power of 0.95, a probability level of 0.05, and an effect size of 0.4. The result showed that 16 participants were needed. By considering potential dropout rates (with an attrition rate of 40%), we recruited 24 participants. The inclusion criteria were as follows: 1) age between 18 and 30 years, and 2) a habit of regular exercise (i.e., aerobic exercise for 30 min once or twice a week). The exclusion criteria were as follows: 1) a history of lower extremity injuries in the preceding 6 months, 2) currently using neuropsychiatric medication, 3) diagnosis of overt neurological diseases, and 4) any contraindications to the use of HD-tDCS (e.g., metal-implanted devices in the brain). Written informed consent was obtained from each participant prior to their participation. The study protocol was approved by the Institutional Review Board of the Shanghai University of Sport (No. 102772022RT050) and conducted according to the Declaration of Helsinki.

In this randomized, double-blinded, and sham-controlled parallel study, 24 participants were randomly assigned to HD-tDCS and control groups (n = 12 in each group) using computer-generated random numbers. The randomization was conducted by an independent researcher who was not involved in the data collection or analysis. Participants underwent five consecutive days of intervention, in which the participants in the HD-tDCS group received HD-tDCS, and those in the control group received sham stimulation. Multiple-sessions of tDCS induce cumulative and longer-term effects on the functions [e.g., cortical neural plasticity changes (Talsma et al., 2017)]. Previous study showed that five consecutive days of tDCS can induce significant effects on CMFC (Padalino et al., 2021), one of the primary outcomes of our study. We thus designed the protocol here by using five consecutive sessions of tDCS. Three days before and within 1 day after the intervention, the participants completed a prolonged running exercise program (Figure 1). Post-fatigue assessments were completed within 5 min after running to avoid rapid elimination of fatigue. The participants were required to maintain their daily lifestyles throughout the study, avoid vigorous activities for 24 h prior to each exercise session, and abstain from caffeine intake 12 h before each exercise session. Participants were allowed to maintain their regular running habits but were instructed to reduce the intensity and duration of their exercise to avoid inducing fatigue. The running velocity and duration were kept consistent for both the two exercises (i.e., the same fatigue degree across two visits).

The participants were instructed to complete a running test on a treadmill (Gymrol S2500, HEF Techmachine, Andrézieux-Bouthéon, France) with a 1% slope to determine individualized constant velocity in prolonged running (Racinais et al., 2007). The velocity was set at 8 km/h for the initial 3 min and then increased with 0.5 km/h per minute until the first ventilatory threshold occurred (Racinais et al., 2007), as measured by gases expired via a wearable metabolic system (K4b2, Cosmed, Rome, Italy) and determined by following an established reference (Davis, 1985). Running velocity corresponding to the occurrence of the first ventilation threshold was then used as individualized constant velocity in prolonged running exercise.

Then, each participant completed one prolonged running exercise on a treadmill at their own constant running velocity before and after the intervention. Heart rate (HR) and rating of perceived exertion (RPE) were recorded every 4 min. Fatigue condition was determined when participants met all of the following criteria: 1) participants cannot maintain the constant running velocity despite strong verbal encouragement, 2) the RPE rating was at least 19, and 3) their HR reached 90% of their maximum HR (i.e., 220 minus age) (Ramos-Campo et al., 2017; Wang et al., 2021). Blood lactate and maximal voluntary isometric contraction (MVC) of ankle dorsiflexors and plantarflexors were measured before and after running-induced fatigue. No significant differences were observed in the individualized constant running velocity (HD-tDCS group vs. control group: 10.7 ± 1.0 km/h vs. 10.5 ± 1.0 km/h, p = 0.532) and duration time (HD-tDCS group vs. control group: 34.8 ± 5.6 min vs. 36.1 ± 8.6 min, p = 0.678) between the HD-tDCS and control groups.

HD-tDCS was administered by a multichannel NeuroConn^® DC-stimulator system (DC-stimulator MC, Neurocare group, Munich, Germany) connected to five ring-type electrodes (3.5 cm²): one placed on the primary motor cortex (M1) as an anode electrode (i.e., Cz) and the remaining four placed around it as return electrodes (i.e., C3, C4, Fz, and Pz) based on the international standard 10/20 system (Figure 2) (Xiao et al., 2022). The Cz region usually covers bilateral lower-limb area of M1 (Mizuno and Aramaki, 2017). Studies have shown that tDCS targeting the Cz region can modulate the activity of M1, thereby influencing motor control and coordination (Park et al., 2019; Hou et al., 2022). The HD-tDCS consisted of 20-min stimulation of current intensity of 2 mA (Workman et al., 2020a; Workman et al., 2020b). The current ramped up to maximum intensity in the first 30 s and ramped down to 0 in the last 30 s of the stimulation. The sham stimulation consisted of 60-s current delivery only (ramping up and down 30 s each at the beginning without supplying current in the remaining period (Xiao et al., 2022). The current intensity and duration length of one stimulation session was determined by considering both safety and maximizing the effects of stimulation. Previous studies showed that 2 mA was the maximum current intensity of one electrode that can safely stimulate the brain without causing serious adverse event (Bikson et al., 2016; Buchanan et al., 2023). Another study showed that the effects of tDCS on neural plasticity reached maximum when stimulating for 20 min, and increasing the duration length of one stimulation session beyond 20 min cannot induce greater effects (Mosayebi Samani et al., 2019). Therefore, almost all the studies using tDCS used this kind of design and we here also followed this design. We here used traditional in-active sham that was believed to not induce potential significant effects on the excitability of targeting regions. The device was operated by a trained research assistant who did not participate in any other procedures of the study. The participants and researchers were unaware of the stimulation condition. Blinding efficacy and adverse effects were assessed at the last stimulation by asking the participants to complete questionnaires to subjectively judge the type of stimulation and report if any side effects or uncomfortable feelings were experienced, respectively (Valiengo et al., 2020).

EEG and sEMG data were pre-processed and calculated using MATLAB R2018b (MathWorks, Inc., Natick, MA, United States). Channels with data of obvious artifacts were rejected by visual inspection. The outliers were interpolated, and the data were re-referenced to an average reference (Forman et al., 2022). Band-pass filter was applied within the frequency range of 0.5–60 Hz (Lu and Sorooshyari, 2023), and independent component analysis was run to remove noises from the eyes, muscles, and heart.

sEMG data were bandpass filtered within the frequency range of 10–500 Hz. The root mean square (RMS) of each muscle was calculated based on the envelope of sEMG signals (Wang et al., 2022). Higher RMS values represent a higher level of muscle activation. The RMS of each muscle was normalized using the peak RMS amplitude of the respective sEMG signals recorded during the ankle dorsi-plantarflexion after the first prolonged running exercise.

The epochs of dorsiflexion and plantarflexion were separated according to the torque curves collected during the ankle dorsi–plantarflexion task.

The linear coupling between synchronously recorded EEG and sEMG signals was analyzed using CMC (Yang et al., 2009; Siemionow et al., 2010; Tuncel et al., 2010; Xu et al., 2023). We here focused on the coupling in the beta (16–30 Hz) and gamma (31–50 Hz) bands as evidence has shown strong coherence between the EEG and sEMG signals within these bands and were associated with motor control (Chakarov et al., 2009; Yang et al., 2009). Pre-processed EEG and sEMG data from the two epochs of dorsiflexion and plantarflexion movements were used to obtain signals in the beta and gamma bands. Then, the CMC of the EEG data of each channel (C1, C2, Cz, FCz) and the sEMG data of dorsiflexor (i.e., TA) and plantarflexors (i.e., MG and LG) in these two bands were calculated as follows (Mima and Hallett, 1999): C x y = S x y f 2 S x x f × S y y f S x y f = 1 n ∑ i = 1 n x i f × y i f * where x and y represent two types of signals, S _xx (f) and S _yy (f) represent the auto-spectra of the signals x and y at frequency f, respectively, and S _xy (f) represents the cross-spectrum of the signal x and y at frequency f. The CMC value is denoted by [0,1], where 0 represents the lack of coherence between the two signals; values closer to 1 indicate a higher degree of synchronization between the signals. CMC was considered significant if it is above the confidence level (CL), which was calculated as follows (Halliday et al., 1995): C L = 1 − 1 − α 1 N − 1 where N represents the number of data segments, and α is the significance level. In this study, the number of data segments was 512, and the significance level was 0.95.

Continuous data were presented as mean ± standard deviation (SD). Shapiro–Wilk test was used to test the normality of all continuous data. When the data were normally distributed, one-way ANOVA models were used to assess significant differences in running performance (i.e., individualized constant running velocity and duration time), physiological metrics (i.e., blood lactate, HR), MVC torques, CMC, and RMS between groups (i.e., HD-tDCS and sham) before the intervention. When the data were not normally distributed, non-parametric Friedman test was used. Wilcoxon signed-rank test was performed to assess significant differences in RPE (non-continuous data) between the two groups before the intervention.

To examine the effects of fatigue on the physiological metrics and MVC torques within the first prolonged running exercise, one-way ANOVA models were used when the data were normally distributed. Otherwise, non-parametric Friedman test was implemented. The dependent variables were blood lactate, HR, and MVC torque for each model, and the factor was time (pre-, post-fatigue). Wilcoxon signed-rank test was performed to determine the effects of fatigue on RPE in the first prolonged running exercise.

To examine the effects of HD-tDCS on CMC and muscular activities of dorsiflexors and plantarflexors under fatigue, two-way repeated-measures ANOVA models adjusted for age, height, and weight were used when the data were normally distributed. To note, if there is a significant difference in any outcomes before the intervention between the two groups, we included the pre-intervention outcomes into the model. The dependent variables were CMC, RMS, and MVC torque for each model. Model factors were stimulation condition (HD-tDCS, sham), time (pre-, post-intervention), and their interaction. Bonferroni’s test was used for post hoc analysis to determine where the significance was in the ANOVA models. Partial eta-square ( η p 2 and Cohen’s d were reported to the ANOVA models and post hoc analyses as effect size values, respectively. When the data were not normally distributed, non-parametric Friedman test was used.

We further explored the relationship between CMC and muscular activities of dorsiflexors and plantarflexors by using Pearson’s correlation analyses. The association between these outcomes before the intervention, as well as between the percent changes of those with significant changes induced by HD-tDCS, was examined.

All statistical analyses were performed using SPSS 21.0 Software (SPSS Inc., Chicago, IL, the United States). The significance level was set at p < 0.05.

The one-way ANOVA models showed that the MVC torques of dorsiflexors (p < 0.001) and plantarflexors (p < 0.001) significantly decreased and the blood lactate (p < 0.001), HR (p < 0.001), and RPE (p < 0.001) significantly increased after fatigue compared with those before fatigue. This finding suggests that the prolonged running exercise successfully caused fatigue (Figure 3).

In the beta band (Table 2), two-way repeated-measures ANOVA models indicated significant interactions between stimulation condition and time on C1-TA (F _{[1, 22]} = 6.943, p = 0.015, η p 2 = 0.240 ) and Cz-TA (F _{[1, 22]} = 6.301, p = 0.020, η p 2 = 0.223 ). Post-hoc analyses revealed that the CMC values between the EEG signals of C1 and Cz and the sEMG signal of TA significantly increased after the HD-tDCS intervention compared with those at pre-intervention (C1-TA: p = 0.011, Cohen’s d = 0.911, Figure 4A; Cz-TA: p = 0.032, Cohen’s d = 0.767; Figure 4B) in the HD-tDCS group and post-intervention (C1-TA: p = 0.038, Cohen’s d = 0.901; Cz-TA: p = 0.045, Cohen’s d = 0.866) in the control group.

In the gamma band (Table 2), the ANOVA models revealed significant interactions between stimulation condition and time on C1-TA (F _{[1, 22]} = 7.067, p = 0.014, η p 2 = 0.243 ) and Cz-TA (F _{[1, 22]} = 6.103, p = 0.022, η p 2 = 0.217 ). Post-hoc analyses revealed that the CMC values between the EEG signals of C1 and Cz and the sEMG signal of TA significantly increased after the HD-tDCS intervention compared with those at pre-intervention (C1-TA: p = 0.023, Cohen’s d = 0.700, Figure 4A; Cz-TA: p = 0.022, Cohen’s d = 0.938; Figure 4B) in the HD-tDCS group and post-intervention (C1-TA: p = 0.046, Cohen’s d = 0.863; Cz-TA: p = 0.045, Cohen’s d = 0.866) in the control group. Additionally, significant main effects of time were observed on Cz-MG (F _{[1, 22]} = 5.867, p = 0.024, η p 2 = 0.211 ), C1-LG (F _{[1, 22]} = 5.565, p = 0.028, η p 2 = 0.202 ), and Cz-LG (F _{[1, 22]} = 7.654, p = 0.011, η p 2 = 0.258 ). This finding indicates a significant decrease from the pre- to the post-intervention values regardless of the group. The other metrics tested had no significant differences (p > 0.055).

The two-way repeated-measures ANOVA models indicated a significant interaction between stimulation condition and time on the RMS value of TA (F _{[1, 22]} = 9.820, p = 0.005, η p 2 = 0.309 , Table 3). Post-hoc analysis indicated that the RMS value of TA significantly increased after the HD-tDCS intervention compared with that at pre-intervention (p = 0.022, Cohen’s d = 0.757) in the HD-tDCS group and post-intervention (p = 0.042, Cohen’s d = 0.880) in the control group. Additionally, significant main effects of time were observed on the RMS values of LG (F _{[1, 22]} = 16.377, p = 0.001, η p 2 = 0.427 ) and MG (F _{[1, 22]} = 4.619, p = 0.043, η p 2 = 0.174 ), reflecting a significant decrease from the pre- to the post-intervention values regardless of the group.

The ANOVA models revealed a significant interaction (F _{[1, 22]} = 7.169, p = 0.014, η p 2 = 0.246 ) between stimulation condition and time on the MVC torque of dorsiflexors (Table 3). Post-hoc analysis revealed that the MVC torque of dorsiflexors significantly increased after the HD-tDCS intervention compared with that at pre-intervention (p = 0.018, Cohen’s d = 0.522) in the HD-tDCS group and post-intervention (p = 0.002, Cohen’s d = 1.395) in the control group. A main effect (F _{[1, 22]} = 8.314, p = 0.009, η p 2 = 0.274 ) of stimulation condition was also observed. Additionally, within the HD-tDCS group, no significant correlation appeared between the percent changes in CMFC and muscular activities of dorsiflexors and plantarflexors after the intervention (p > 0.308).