Twenty-one healthy young adults from Zhejiang University were recruited in this study via an online advertisement. Among them, two subjects were excluded due to data acquisition issues, and the remaining 19 subjects (male/female = 13/6, age = 22.16 ± 0.65 years) were included for the following analysis. All participants had normal or corrected-to-normal vision. None of them reported cardiovascular, cognitive disorder, color blindness, or any acute illness that might influence the cognitive task performance. Each participant was informed of the experimental protocol and potential risks through a telephone interview. They were required to get a full night of sleep (> 7 h, recorded by HUAWEI-B19 bracelet) for two continuous nights and were instructed to abstain from caffeine and alcohol and avoid vigorous physical activity for 7 h before the experiment. Those participants who failed to meet the requirements were either ruled out from the experiment or rescheduled. Upon arriving at the lab, written informed consent and Physical Activity Readiness Scale (PARS) were provided to exclude any contraindication during exercise. This study was approved by the Institutional Review Board of Zhejiang University and was carried out in compliance with the Declaration of Helsinki.

A within-subject design was adopted in this work to explore the effects of receiving different types of breaks on mental fatigue recovery, where each participant was requested to finish three sessions in a counterbalanced manner with approximate 1-week interval (Figure 1). During each session (namely, No-Break, Exercise-Break, and Rest-Break session, respectively), participants were required to perform a 30-min PVT with or without a break. The design of PVT was the same as in our previous studies (Gao et al., 2020). Briefly, during the test, subjects were required to monitor a red dot that appeared on the screen and responded to this stimulus by pressing the space bar as soon as possible. Intervals between trials varied randomly from 2 to 10 s (mean = 6 s).

In the No-Break session, participants were required to perform the PVT task continuously without a break, while during the Rest- or Exercise-Break session, participants were required to perform PVT tasks with a 15-min still sitting with eyes fixed on a black cross (namely mid-task rest) or cycling on a Monark 975 stationary exercise bicycle (namely mid-task exercise) in the middle of the tasks. Sessions were arranged at the same time of the day to eliminate the impact of circadian rhythm (Pontifex et al., 2019). The mid-task rest was introduced with the notion that rest breaks induced mental fatigue recovery (Lim et al., 2013; Sun et al., 2017; Qi et al., 2020), while the mid-task exercise corresponded to the view that the moderate-intensity exercise led to cognitive improvement (Chang et al., 2012). The intensity of the cycling break was estimated at individual-level using the following formula:

where HR stands for heart rate, and HR_max was set as 220 – age, HR_rest was obtained after participants resting for 5~7 min (recorded by the Polar Vantage V). In addition, subjective feelings of mental states were assessed via the Short Stress State Questionnaire (SSSQ) (Helton et al., 2005; Lim et al., 2016; Qi et al., 2020) before and immediately after each session. Three minutes was determined as the length of an epoch to better illustrate the transient effect of post-break while including enough PVT trials for a stable behavioral and electrophysiological metrics estimation.

High-density continuous EEG was recorded from 64 Ag/AgCl scalp electrodes (model: Brain Products MR Plus, Germany) according to the International 10–20 system during the 30-min PVT for three sessions. The sample rate was set at 2,000 Hz with reference to the FCz. The impedance was kept below 5 kΩ during the whole data collection. In the offline analysis, a previously validated standard EEG preprocessing pipeline was adopted. Briefly, all data were down-sampled to 256 Hz, bandpass filtered into 1–40 Hz, 50 Hz notch filtered, and average re-referenced. Independent component analysis (ICA) (Jung et al., 1997) was further used to detect and eliminate artifacts (i.e., EOG and eye movements). Then for each 3-min epoch, the obtained EEG data were de-trended and segmented into 6-s trials (corresponding to the average time of one behavioral trial in the PVT). EEG data in an epoch with a voltage >100 uV was rejected (Zhao et al., 2012). For all preprocessing steps, customized codes and the EEGLab toolbox (Delorme and Makeig, 2004) were used in MATLAB 2020B.

Each 6-s EEG data trial was divided into segments using a 100% Hamming window of 1 s, then the power spectral density (PSD) was estimated by Welch's method, and specially focused on two low-frequency bands: delta band (1–4 Hz) and theta band (4–7 Hz) for their potential sensitivity toward mental states (Tran et al., 2020). For each channel, relative power (P_δ and P_θ) was estimated as the ratio of the specific band power to the total power spectral that was estimated with a frequency range of 1–40 Hz. For each 3-min epoch, averaged relative power was obtained from all included trials.

Meta-analysis suggested that spectral EEG across frontal, central, and posterior cortical regions was related to mental fatigue, and the effect size was large for the central region and moderate for the frontal and posterior regions (Tran et al., 2020). Considering these region-dependent alterations of EEG power during TOT (Craig et al., 2012; Lim et al., 2013), three bilateral clusters (corresponding to frontal, central, and parietal areas) were divided to better investigate the restorative effect of different breaks. The assignment of electrodes in each cluster is displayed in Figure 2.

To investigate the time-on-task (TOT) effect induced by PVT, we averaged the reaction time (RT) in the No-Break session into 10 3-min bins to evaluate the trend of behavioral performance over the task (Figure 3A, gray curve). As expected, performing PVT led to a significant TOT effect, manifesting increments in RT similar to our previous studies (Sun et al., 2017). Specifically, a significant increase [F_{(3, 54)} = 22.270, p < 0.001, η² = 0.553] of reaction time was observed from T1 to T4. The individually fitted linear slope of the RT also showed a performance decline throughout the task (slope Mean ± SD = 0.022 ± 0.004), in which averaged RT was 11.78% longer during T4 (most fatigued) in comparison with T1 (most vigilant).

For the immediate effect (T2 vs. T3) of mid-task breaks (Figure 3B), there were significant main effects for time-by-session interaction [F_{(2, 36)} = 4.942, p = 0.013, η² = 0.215], time [F_{(1, 18)} = 25.713, p < 0.001, η² = 0.588], and session [F_{(2, 36)} = 7.421, p = 0.002, η² = 0.292]. Following post-hoc analyses indicated that the significant interaction effect was attributed to the decrease of RT in both the Exercise-Break session [F(1, 18) = 16.448, p = 0.001, η² = 0.477] and the Rest-Break session [F_{(1, 18)} = 17.838, p = 0.001, η² = 0.498]. However, no significant difference was observed between the Exercise-Break session and Rest-Break session in their post-break RT, indicating a similar vigilance improvement immediately after different interventions.

The general effect (T1 vs. T4) of mid-task breaks was also tested (Figure 3C). Significant main effects were found for time [F_{(1, 18)} = 65.451, p < 0.001, η² = 0.784] and sessions [F_{(2, 36)} = 4.655, p = 0.016, η² = 0.205], but no time-by-session interaction [F_{(2, 36)} = 2.059, p = 0.142, η² = 0.103]. Follow-up post-hoc comparisons for the main time effect suggested that RT in T4 was significantly longer than that in T1 (p < 0.001 in all sessions). Specifically, compared with T1, RTs in Exercise-Break and Rest-Break sessions increased by 6.970 and 8.297%, respectively. In addition, post-hoc comparisons for the main session effect indicated that in the T4, the RT in the Exercise-Break session was significantly shorter than that in the No-Break session [F_{(1, 18)} = 8.431, p = 0.009, η² = 0.319].

In terms of subjective states, significant main effect of time was found on both engagement [F_{(1, 18)} = 21.535, p < 0.001, η² = 0.545] and distress [F_{(1, 18)} = 8.568, p = 0.009, η² = 0.322] scores, but no significant difference of session [engagement: F_{(2, 36)} = 1.032, p = 0.367, η² = 0.054; distress: F_{(2, 36)} = 1.762, p = 0.186, η² = 0.089] and time-by-session [engagement: F_{(2, 36)} = 0.804, p = 0.455, η² = 0.043; distress: F_{(2, 36)} = 0.113, p = 0.894, η² = 0.006] interaction. Further inspection of the main effect in time suggested that for No-Break and Exercise-Break session, participants showed significantly increased distress [No-Break session: F_{(1, 18)} = 13.986, p = 0.001, η² = 0.437; Exercise-Break session: F_{(1, 18)} = 6.495, p = 0.02, η² = 0.265] as well as less engagement [No-Break session: F_{(1, 18)} = 17.618, p = 0.001, η² = 0.495; Exercise-Break session: F_{(1, 18)} = 4.015, p = 0.06, η² = 0.182] in the post-session compared with pre-session, indicating a similar trend of subjective experience between successive PVT and exercise-inserted PVT.

The immediate effect of mid-task breaks on the EEG power ratio was first assessed. We observed a significant difference in theta and delta activities in T3 (post-break bin) compared with T2 (pre-break bin). Specifically, in theta band, there were significant main effects for time-by-session interaction, in P_{θ_Frontal} [F_{(2, 36)} = 5.848, p = 0.006^*, η² = 0.245; ^* indicated survive FDR threshold at q < 0.05 for interaction effect analysis] and P_{θ_Parietal} [F_{(2, 36)} = 3.886, p = 0.030^*, η² = 0.178]. Following post-hoc analysis indicated that, in the Exercise-Break session, P_θ significantly decreased in T3 compared with T2, in P_{θ_Frontal} [F_{(1, 18)} = 11.528, p= 0.003, η² = 0.390], P_{θ_Parietal} [F_{(1, 18)} = 21.381, p < 0.001, η² = 0.543] in comparison with a maintained P_θ in Rest-Break and No-Break sessions (Figure 4).

And in delta band, there were significant main effects for time-by-session interaction, in P_{δ_Central} [F_{(2, 36)} = 3.443, p = 0.043, η² = 0.161]. The post-hoc analysis indicated that, in the Exercise-Break session, P_{δ_Central} significantly increased in T3 compared with T2 [F_{(1, 18)} = 4.676, p = 0.044, η² = 0.206] (Figure 5).

We assessed the general effect of mid-task breaks by comparing the EEG power ratio in T4 (last 3-min bin of the whole session) with T1 (first 3-min bin). A significant main effect for time-by-session interaction was also observed in P_{δ_Central} [F_{(2, 36)} = 5.221, p = 0.010^*, η² = 0.225]. The post-hoc comparison showed a significant decrease in T4 compared to T1 [F_{(1, 18)} = 6.171, p = 0.023, η² = 0.255] in the Rest-Break session, as well as an increase trend [F_{(1, 18)} = 3.212, p = 0.090, η² = 0.151] in the Exercise-Break session (Figure 6).

According to the analysis of immediate and general effect, P_δ and P_θ demonstrated significant differences immediately after the mid-task exercise, while P_δ also suggested a significant difference in the general effect of the Rest-Break session. Consistent with the interaction effect observed in EEG power, correlation analyses were mainly focused on the immediate effect of the Exercise-Break session and the general effect of the Rest-Break session. Correlation investigation between the EEG power and behavioral performance showed that P_θ in T2 was correlated with ΔRT_T3−T2 (r₁₉ = −0.502, p= 0.028) in the exercise-inserted task, and P_δ in T3 (r₁₉ = 0.537, p = 0.018) was significantly related with ΔRT_T4−T1 in the rest-intervention task. No significant association was observed between ΔP_θT3−T2 and ΔRT_T3−T2 (r₁₉ = 0.124, p = 0.613), ΔP_δT3−T2 and ΔRT_T3−T2 (r₁₉ = −0.055, p = 0.822) in Exercise-Break session, as well as ΔP_δT4−T1 and ΔRT_T4−T1 (r₁₉ = −0.425, p = 0.070) in Rest-Break session. In particular, the EEG power in correlation analysis was acquired from global electrodes rather than specific clusters to preliminarily explore the biomarker of mental fatigue recovery (Figure 7).

Both the mid-task exercise and rest induced similar immediate improvements as shown in the significantly reduced RT in the post-break window. For the mid-task rest, this behavioral result was inconsistent with our early findings that no significant restorative effect was observed after rest breaks (Sun et al., 2017). This discrepancy could stem from the different duration of the rest admitted to the participants, i.e., 5 min in Sun et al. (2017) vs. 15 min in the current work. Embracing previous observations with the current findings, we could therefore infer that a longer rest break might be needed to induce substantial recovery from mental fatigue (Ross et al., 2014; Helton and Russell, 2015). Further evidence to support this notion was from a behavioral study quantitatively investigating the effect of rest in three different durations for TOT recovery, which demonstrated that longer rest break produced a better restoration (Lim and Kwok, 2015). However, no significant recovery was observed at the end of the Rest-Break session in our results. This is in line with the previous study that TOT decrements tended to be steeper after long than short breaks (Lim and Kwok, 2015). As a result, providing participants with a mid-task rest with the time duration of 15 min might only produce a transient restoration restricted to the period right after the midway break.

The mid-task exercise breaks were found to be capable of replenishing the behavioral performance immediately, fairly as much as rest did. Notably, these results were consistent with accumulating evidence that aerobic exercise produced a more robust positive effect on cognitive and attention (Pontifex et al., 2019). This was also partly consistent with the arousal theory, which indicated that exercise-induced brain activation, especially in the prefrontal cortex (Popovich and Staines, 2015), the region critical for sustained attention. Our work extends the above findings by linking the positive effects of exercise to the practical application of mental fatigue restoration.

More interestingly, we found the subjective distress score in SSSQ, a measure of negative affect, kept consistent in the Rest-Break session, while increased in the No-Break session and Exercise-Break session. This indicated that after the 15-min cycling break, subjects might not feel recovered from mental fatigue, even though there were indeed apparent improvements in their vigilance. In addition, the trend of engagement scores in SSSQ also reflected a similar phenomenon. Put differently, although both rest and exercise could intervene TOT effect immediately, subjects might have different recognition of their “feeling about mental fatigue,” which might correspond to divergent cognitive processes underlying the similar behavioral performance improvement.

As for the mid-task exercise, a significant reduction of theta activity in the frontal and parietal cortex areas was observed in the immediate post-break period. Frontal theta has been repeatedly reported as a reliable index for mental fatigue (Lal and Craig, 2001; Wascher et al., 2014), and an increase in frontal theta may reflect exhausting cognitive control resources with increasing TOT (Clayton et al., 2015). For example, studies found that reduced frontal theta was associated with better performance in action execution, while higher frontal theta might reflect an excessive consumption of attention engagement and hamper the task performance (Kao et al., 2013). Besides, a significant theta power increase in the frontal was believed to reflect a more demanding attention control process that originated in the anterior cingulate cortex (ACC) (Sauseng et al., 2010). This ACC-mediated theta activity has been believed to be associated with externally oriented monitoring and executive control (Cavanagh and Frank, 2014; Cohen, 2017). From the correlation analysis, we further observed that a greater before-intervention P_θ was associated with more behavioral restoration after mid-task exercise in our study. In summary, the reductions in theta following exercise break might reflect optimized monitoring and allocation of executive control resources toward external goals so as to ameliorate TOT declines. In other words, the mid-task exercise matched the current attention to the desired level.

Besides, a significant increase in the central delta was also observed right after exercise intervention. Consistent with our results, the delta increase has been repeatedly reported immediately after exercise (Mechau et al., 1998; Crabbe and Dishman, 2004). The traditional view toward the increase of delta recognized it as the biomarker of fatigue. For instance, in laboratory-controlled driving simulating studies, increases in delta activity were found during the transition into the fatigued state (Lal and Craig, 2002). Unlike these findings, an increased delta ratio was observed in the current work where significant improvements in behavioral performance were revealed after cycling exercise, which might suggest a beneficial effect of the increased post-exercise delta. Of note, in addition to fatigue-related alterations, the increased delta power has also been reported to be associated with inhibition of the internal interferences that might affect the task performance (Harmony, 2013), which may lead to the observed increased post-exercise delta power. For example, during a Go/No-Go task in 15 normal young volunteers, delta activity increases were clearly demonstrated while subjects inhibited their movement (Harmony et al., 2009). Besides, another research suggested that there was a reciprocal relationship between alpha and delta activity, which reflected the inhibitory control over motivational and emotional drives (Knyazev, 2007). In summary, we inferred that the delta increases after exercise suggested an inhibitory effort toward possible interference thoughts, which in turn improved the performance and represented fatigue recovery.

In sum, these findings could fit some of the predictions in the oscillatory frequencies model of sustained attention (Clayton et al., 2015), in which attention is adjusted through the excitation of task-relevant cognitive processes and the inhibition of task-irrelevant cognitive processes. The decrease of theta after exercise could demonstrate a more effective cognitive monitoring and manipulation in primary sustained attention tasks. And the increase of delta might reflect the need for inhibiting task-irrelevant thoughts during a primary mental task. The two approaches together could promote the performance of post-break behavioral performance.

In contrast with the post-exercise effect, EEG change was not observed immediately after rest intervention in low-frequency power spectral but manifested as a drop in delta power at the end of the task when the TOT effect reappeared. In addition to the biomarker of mental fatigue and internal inhibition, delta power was also considered associated with negative effects (Wu et al., 2019; Zheng et al., 2019) and subjective sleepiness (Phipps-Nelson et al., 2011). Therefore, the decrease of the delta at the end of the Rest-Break Session could be illustrated as the active emotion and motivation toward sustained attention task, so that participants would respond positively when mental fatigue was induced once again. This beneficial effect brought by mid-task rest could also be validated from the subjective feeling feedback, in which subjects maintained the effect and motivation from the start to the end of the Rest-Break Session, while both the No-Break Session and Exercise-Break Session suggested increased subjective distress and declined task engagement. In addition, we found that the level of delta power following rest was associated with long-term behavioral performance based on correlation analysis. In detail, a lower delta power after rest corresponded to a higher sustained attention level before the end of the task, and the delta could be used as a neurological indicator demonstrating the vigilance level after rest. In sum, mid-task rest improved the subjective experience to counter TOT effects, which could be manifested in the delta power after rest intervention.

Some issues should be considered when interpreting our results. First, a within-subject design was adopted in this study to neutralize the widely revealed between-subject variability toward the TOT effect (Parasuraman and Jiang, 2012). Although this is an advantage of the study design, our sample size was subsequently small, and only healthy and young participants were recruited given the amounts of labor in repeated measure (i.e., each participant should come to the lab once a week for continuous 3 weeks without dropout). Therefore, more experimental studies are needed to further validate our findings with a broad age range and a larger independent sample. Besides, subjects were instructed to keep their eyes open with gaze at a central fixation cross during the mid-task rest. It is a more practical way of performing the experiment without the need to give a cue to tell subjects to open their eyes or if concern of falling asleep during the experiment that might induce additional confounding factors compared to a natural eye-closed rest. In fact, a study of multi-center neuroimaging data showed that data collection with eyes closed showed more sleep than the fixation group, suggesting that fixation supports the maintenance of wakefulness (Tagliazucchi and Laufs, 2014). Nonetheless, in a real-life situation, people intend to take a break in an ecologically valid way, with an unconstrained activity that is more relaxing. Therefore, future studies could pay more attention to the naturality and validity of the mid-task breaks to explore the practical restoration approach for a particular application scenario. Second, to induce significant performance decrement, the simple and monotonous PVT task was introduced in this study, which was considered to be more stressful and effortful than complex tasks (Langner and Eickhoff, 2013). In fact, handling multiple tasks were common and important in daily life, with a series of decision-making, alertness, or physical movement, corresponding to more sophisticated activating states of the brain. Therefore, future work could consider the paradigm design consisting of simulated real-world working assignments to further verify the effectiveness of the divergent fatigue-intervention strategy. Third, based on the cortical oscillation model (Clayton et al., 2015), we found that the excitation of task-relevant cognitive processes and the inhibition of task-irrelevant processes could explain the EEG change in theta and delta following mid-task exercise. It is noteworthy mentioning that we did not directly distinguish the task-irrelevant cognitive state from the performance of external stimuli response in the current work. As a result, future research may especially focus on the brain status after break intervention through the neurofeedback technique to further investigate our novel interpretation.