Shoichi Asaoka, Department of Somnology, Tokyo Medical University 6-7-1, Nishishinjuku, Shinjuku-ku, Tokyo 160-0023, Japan. Tel.: +81 3 3342 6111 (ext 5757); fax: +81 3 3342 7083; e-mail: email@example.com
Performance monitoring is an essential function involved in the correction of errors. Deterioration of this function may result in serious accidents. This function is reflected in two event-related potential (ERP) components that occur after erroneous responses, specifically the error-related negativity/error negativity (ERN/Ne) and error positivity (Pe). The ERN/Ne is thought to be associated with error detection, while the Pe is thought to reflect motivational significance or recognition of errors. Using these ERP components, some studies have shown that sleepiness resulting from extended wakefulness may cause a decline in error-monitoring function. However, the effects of sleep inertia have not yet been explored. In this study, we examined the effects of sleep inertia immediately after a 1-h daytime nap on error-monitoring function as expressed through the ERN/Ne and Pe. Nine healthy young adults participated in two different experimental conditions (nap and rest). Participants performed the arrow-orientation task before and immediately after a 1-h nap or rest period. Immediately after the nap, participants reported an increased effort to perform the task and tended to estimate their performance as better, despite no objective difference in actual performance between the two conditions. ERN/Ne amplitude showed no difference between the conditions; however, the amplitude of the Pe was reduced following the nap. These results suggest that individuals can detect their own error responses, but the motivational significance ascribed to these errors might be diminished during the sleep inertia experienced after a 1-h nap. This decline might lead to overestimation of their performance.
Accidents and sleepiness are closely related. Approximately 9–20% of vehicle accidents are thought to be sleep related (Horne and Reyner, 1995; Maycock, 1996). The relative frequency of these accidents peaks in early morning and mid afternoon, which are synchronized to points in the circadian cycle associated with sleepiness (e.g. Garbarino et al., 2001; Horne and Reyner, 1995; Maycock, 1996). This phenomenon is also observed in the patterns of errors made by individuals reading gas meters (Mitler and Miller, 1996). To date, many laboratory-based studies have demonstrated that sleepiness resulting from acute or cumulative sleep deprivation is associated with impairments across a wide variety of cognitive functions (see Durmer and Dinges, 2005 for review). Recently, not only simple cognitive performance, but also higher cognitive functions such as decision making (Harrison and Horne, 2000) or self-monitoring of performance (Dorrian et al., 2003), have been shown to be impaired by sleepiness.
There has been limited research involving the relationship among the ERN/Ne, Pe and sleepiness, and even these few studies have produced mixed results. Scheffers et al. (1999) reported that ERN/Ne amplitude is reduced after one night without sleep. Tsai et al. (2005) also showed that ERN/Ne amplitude was reduced after 26 h of sleep deprivation. Tsai et al. (2005) also reported a reduction in Pe amplitude compared with those measured after normal sleep. However, in another experiment conducted by the same lab (also 26 h of sleep deprivation) they reported that the Pe was not reduced, when subjects were instructed to correct any observed errors (Hsieh et al., 2007). In contrast, Murphy et al. (2006) used a more moderate amount of sleep deprivation (20 h) and found no significant reduction in ERN/Ne amplitude. However, they did report a reduction in Pe amplitude. Although the effect of sleepiness caused by extended wakefulness on the ERN/Ne and Pe has not been consistent among these various experiments, taken together these results strongly suggest deterioration of the error-monitoring function during sleepiness after 20–26 h of sleep deprivation.
However, sleepiness can be caused by processes other than extended wakefulness. There is a period of time immediately after sleep offset where there is some residual sleepiness. This transient period from sleep to fully awake has been termed ‘sleep inertia’. The results of previous studies suggest that subjective sleepiness may be severe (e.g. Hayashi et al., 2003a,b; Jewett et al., 1999) and cognitive performance is impaired during sleep inertia (e.g. Hofer-Tinguely et al., 2005; Jewett et al., 1999; Stampi et al., 1990). Moreover, similar to the effects of sleep deprivation, during sleep inertia, more complex processing (e.g. decision making) is also disturbed (Bruck and Pisani, 1999).
It is well known that the intensity of sleep inertia depends on the sleep stage which individuals are awakened from. Individuals aroused from slow-wave sleep (SWS) report more severe sleep inertia (e.g. Dinges et al., 1985; Stampi et al., 1990). During a daytime nap, SWS typically begins about 15 min after sleep onset (Fushimi and Hayashi, 2008). Therefore, longer naps (>15 min) should induce stronger sleep inertia compared with a shorter nap (<15 min). This method of studying sleep inertia after a nap has been employed several times; however, there are no studies that explored the effects of sleep inertia on error detection processes. Thus, the present study examined the effects of sleep inertia on error detection processes immediately after a 1-h daytime nap, using the ERN/Ne and Pe as markers of these cognitive functions.
Materials and methods
Nine healthy young adults with no self-reported sleep problems and no habitual daytime napping participated in this study (mean age 24.0 ± 1.1 years, seven male). Written informed consent was obtained from each participant before participation. Participants completing the entire protocol were paid an honorarium. The data of one male were excluded from all analyses mentioned below, because he made too few errors to obtain a stable average of the error-related ERP waveform.
Participants completed both nap and rest conditions in a repeated-measures design. They were instructed to obtain 7 h sleep from 0:00 to 07:00 hours and no other sleep during the 3 days before the experiment. In addition, on the day of each experimental condition and the previous day they were prohibited from caffeine and alcohol consumption. Excessive physical exercise on the experimental days was also prohibited. Participants recorded a sleep log for 5 days prior to each experimental day and also wore an actigraph (Actiwatch-L®; Mini Mitter/Respironics, Bend, OR, USA) for 3 days before the experiments to confirm their sleep–wake pattern. There was a minimum of 5 days between conditions. In both conditions, the first cognitive task session started at 13:00 hours in an air-conditioned soundproof chamber. In the nap condition, participants went to bed at 14:00 hours, and they were awakened at 15:00 hours. Ninety seconds after their sleep offset, the second cognitive task session started. In the rest condition, they had 1-h rest in the chamber instead of a nap, followed by the testing session at 15:00 hours. The order of the two conditions was counterbalanced among the participants. Participants were prohibited from exiting the experimental chamber from the start of the first cognitive task session to the end of the second session, except for going to the lavatory. They were allowed to read magazines, books, comics and newspapers during their 1-h rest period and between the end of the first cognitive task session and the start of a 1-h nap or rest period. The brightness in the chamber was maintained at 25 lx at the participants’ eye level, except during the nap when it was reduced to 0 lx. After each cognitive task session, participants reported their subjective estimation of alertness, sadness, tension, loss of motivation, happiness, weariness, calmness, and sleepiness, as well as task performance, effort produced on each task and error rate using Visual Analog Scales (VAS; Monk, 1989). The procedures of this study were approved by the ethics committee of Waseda University.
We tested a stimulus–response compatibility task referred to as the arrow-orientation (AO) task (Masaki and Segalowitz, 2004). According to Kornblum’s (1992) taxonomy, the AO task is in the Type 8 ensemble (e.g. a spatial Stroop task). In this task a white fixation cross (0.7º× 0.7º) was presented for 300 ms in the center with black background on a computer monitor, placed 1 m in front of the participant. Then a white arrow (pointing up or down) was presented above or below fixation with an eccentricity of 0.8° visual angle (between centers of fixation and arrow) for 200 ms (Fig. 1). Thus, stimuli consisted of two compatible and two incompatible stimuli in relation to the direction which the arrow was pointing. The angle subtended by the arrow was 0.7º× 0.4º. The display was then blank for 500 ms until the next fixation point was presented. The task was to respond to the pointing direction of the arrow (up or down), but not to the location (above or below), by pressing a button with the corresponding hand. Both speed and accuracy were emphasized during task instructions. To familiarize participants with the task, a practice session (40 trials × four blocks) was performed 3–5 days before the first experimental day. Each experimental session consisted of four blocks of 300 trials each. Participants had a 1-min rest between each block. They also practiced the task (12 trials) before each session. Each experimental session lasted approximately 25 min.
The electroencephalogram (EEG) was recorded from FCz, C3, C4, Pz, Oz, A1 (left earlobe) and A2 (right earlobe) with Ag/AgCl electrodes using average reference of all electrodes. Horizontal electrooculograms (hEOG) were recorded from the left and right outer canthi, and vertical electrooculograms (vEOG) from above and below the left eye. In the nap condition, the chin electromyogram was also bipolarly recorded. These were recorded with DC and 140 Hz low-passed filter, using the QuickAmp® (Brain Products GmbH, Munich, Germany). All electrode impedances were kept below 10 kΩ. All physiological signals were digitized at a rate of 500 Hz.
Behavioral measures and event-related potential analysis
For each session, reaction time (RT) for correct trials and number of error responses were computed. These variables were calculated separately according to stimulus compatibility and condition. EEG was re-referenced offline to averaged earlobes. EEG artifacts due to eye blinks were corrected using the method outlined by Gratton et al. (1983). There were relatively fewer errors to compatible stimuli; therefore, we could not obtain stable ERP-averaged waveforms for some participants in at least one session. As we employed a within-subjects design, we would lose these participants in any analyzes and the remaining participants would not be sufficient for any meaningful analyzes to be carried out. Therefore, ERPs were calculated using only the segments with incompatible stimuli. We averaged response-locked waveforms with error and correct trials, respectively, and subtracted correct trial averages from error trial averages to clarify the ERN/Ne and Pe waveform for scoring. Finally, ERP waveforms were filtered offline with a bandpass of 0.016–10 Hz. A mean voltage from 600 to 400 ms before the response was used as a baseline to minimize any influence of the stimulus-related P3 (Murphy et al., 2006). The average RTs for incompatible stimuli were approximately 400 ms in this study, thus the period used for the baseline was approximately the time that the fixation was presented on the display. ERN/Ne amplitude was measured as the negative peak in a time window ranging from 50 to 150 ms after the response relative to the preceding positive peak (peak to peak measurement) at FCz. For Pe amplitude, a mean voltage from 200 to 500 ms after the response was measured at Pz.
Sleep stage of the nap was scored offline according to standard criteria (Rechtschaffen and Kales, 1968). Total sleep time, time spent in stages 1–4 and rapid eye movement (REM) were determined, respectively.
Each subjective measure from the VAS and two ERP amplitudes was compared with a repeated-measure two (nap versus rest) × two (before versus after nap/rest) anova. RTs for correct responses and number of errors were compared with repeated three-way anova with the two factors listed above as well as stimulus type (compatible versus incompatible). If any interactions were significant, simple effects analyses were performed. Post hoc analyses were done using the Bonferroni correction.
Sleep schedules and nap architecture
The individual sleep schedules across the two conditions were comparable in terms of bedtime before the experimental day (t7 = 0.80, P =0.45) and rise time on the experimental day (t7 = 0.41, P =0.69). The 1-h nap period consisted of 6.8 ± 2.8 min of stage 1, 32.9 ± 11.5 min of stage 2, 7.1 ± 5.9 min of stage 3, 2.4 ± 4.6 min of stage 4 and 3.1 ± 5.7 min of stage REM. Average total sleep time was 52.3 ± 11.3 min. Every participant had at least one SWS sleep epoch. Three participants were awakened from stage 2, two participants from stage 3, one participant from stage 4 and two participants from stage REM.
Behavioral and subjective measures
In general, participants had fewer errors (F1,7 = 14.80, P <0.01, pη2 = 0.68) and shorter RTs (F1,7 = 38.09, P <0.001, pη2 = 0.85) for compatible stimuli (N of errors: 34.47 ± 8.85; RT: 371.56 ± 9.09 ms) compared with incompatible stimuli (N of errors: 92.28 ± 17.33; RT: 404.97 ± 10.71 ms). The main effects of session and condition were not significant in terms of RT (session, F1,7 = 0.13, NS, pη2 = 0.02; condition, F1,7 = 0.01, NS, pη2 < 0.01) and number of errors (session, F1,7 = 0.86, NS, pη2 = 0.11; condition, F1,7 = 0.37, NS, pη2 = 0.05). There were no significant interactions on RT (condition × session, F1,7 = 0.57, NS, pη2 = 0.08; condition × stimuli, F1,7 = 0.13, NS, pη2 = 0.02; session × stimuli, F1,7 = 2.80, NS, pη2 = 0.29; condition × session × stimuli, F1,7 = 0.01, NS, pη2 < 0.01) and number of errors (condition × session, F1,7 < 0.01, NS, pη2 = 0.01; condition × stimuli, F1,7 = 0.01, NS, pη2 < 0.01; session × stimuli, F1,7 = 0.55, NS, pη2 = 0.07; condition × session × stimuli, F1,7 = 0.49, NS, pη2 = 0.07; Table 1).
Table 1. Arrow-orientation task performance and amplitudes of ERPs
While the interaction of condition × session did not reach significance (F1,7 = 3.93, P =0.09, pη2 = 0.36), participants tended to estimate their performance as better in session 2 compared with session 1 of the nap condition (simple effects, P =0.03; Fig. 2). There was a significant interaction of condition × session on subjective effort (F1,7 = 6.15, P =0.04, pη2 = 0.47). Simple effects analysis showed increased subjective effort during the nap condition compared with the rest condition in session 2 (P <0.01). Also, in the rest condition, participants tended to report more effort after session 1 than after session 2 (P =0.08). In contrast, during the nap condition they tended to report more effort after session 2 than after session 1 (P =0.09). There were no significant main effects or interaction on subjective error rate estimation. While the participants reported more weariness and motivation in session 1 than in session 2 (F1,7 = 7.44, P =0.03, pη2 = 0.52; F1,7 = 9.58, P <0.02, pη2 = 0.58), there were no other significant main effects or interactions on any other subjective scales.
Event-related potential measures
Waveforms of response-locked ERPs for erroneous responses and correct responses (Fig. 3a), and difference waves (Fig. 3b; erroneous–correct response) are shown in Fig. 3. While there were no main effects and no interaction on the amplitude of ERN/Ne at FCz (condition × session: F1,7 = 0.001, P =0.98, pη2 < 0.001), a significant interaction of condition × session was found for the amplitude of the Pe at Pz (F1,7 = 11.78, P =0.01, pη2 = 0.63). Simple effects analysis showed that the Pe amplitude during session 2 of the nap condition was smaller than those on session 1 of the nap condition and session 2 of the rest condition (Ps<0.05; Table 1). Main effects of condition and session were not significant in terms of Pe amplitude.
Although previous studies have shown a deterioration in performance immediately after sleep offset (e.g. Tassi and Muzet, 2000), neither RT nor the number of errors was affected by sleep inertia immediately after a 1-h nap in the present study. This result suggests that the current paradigm may have produced only a relatively mild level of sleep inertia. In their review article, Tassi and Muzet (2000) argued that the severity of sleep inertia is influenced by several factors, such as existence of prior sleep deprivation, time of day of awaking, sleep duration and sleep stage prior to awakening. It is well known that sleep inertia becomes more severe when participants have sleep debt (e.g. Dinges et al., 1985; Tassi et al., 2006), awaken at the time near the trough of core body temperature (i.e. early morning; Dinges et al., 1985; Naitoh et al., 1993) or awaken from deep sleep (Dinges et al., 1985; Stampi et al., 1990). In this study, while some participants awoke from SWS, all of them had 7 h of sleep in the night before the experimental day and arose in the afternoon. Thus, our procedures may have resulted in a relatively weak effect of sleep inertia on performance measures. However, participants also reported increased effort to perform the task during sleep inertia. This result means that more effort was required to maintain a comparable level of cognitive performance during sleep inertia. Similar compensatory phenomena have been reported elsewhere (Dinges and Kribbs, 1991; Szelenberger et al., 2005).
Despite the relatively weak effect of sleep inertia on behavioral measures, Pe amplitude was clearly reduced. While the amplitude was scored based on difference waves, this trend was also observed in the raw grand-averaged waveforms of erroneous responses. These results suggest that Pe might be more sensitive to sleep inertia than behavioral measures, and that even mild sleep inertia may reduce the motivational significance or conscious recognition of errors. The deterioration of Pe during sleepiness was observed in previous studies examining sleep deprivation (Murphy et al., 2006; Tsai et al., 2005); therefore, the results of this study concerning the Pe are consistent with the report that the sleepiness resulting from sleep inertia and sleep deprivation were qualitatively similar (Balkin and Badia, 1988). In addition, the VAS scores showed that participants estimated their performance as better during sleep inertia despite no objective difference in actual performance between the two conditions. These results suggest that the reduced motivational significance of errors experienced during sleep inertia reflected by a reduction in Pe might lead to overestimation of their own task performance.
In contrast, ERN/Ne amplitude did not show any significant difference, and the effect size of interaction of condition × session was very small. This result means that response monitoring might not be impaired by sleep inertia, reconfirming the findings of Murphy et al. (2006), who reported no significant reduction of ERN/Ne amplitude after 20 h of consecutive wakefulness. Thus, it is plausible to conclude that sleepiness might not do harm to error detection or conflict detection, but may dampen recognition or emotional evaluation of error response.
Nevertheless, the current results cannot exclusively confirm independence of response monitoring from sleepiness. Indeed, three studies with one night (24–26 h) of sleep deprivation showed a decline in ERN/Ne amplitude (Hsieh et al., 2007; Scheffers et al., 1999; Tsai et al., 2005). It is plausible that more than 20 h of sleep deprivation is the necessary condition to observe a significant deficit in performance monitoring.
It should be noted that the AO task (classified as a spatial Stroop task; Kornblum, 1992) induces more response conflict than other stimulus–response compatibility tasks (i.e. the flanker task and the Simon task). Masaki and Segalowitz (2004) elucidated that the spatial Stroop task adopted in our study resulted in longer RT and more errors than the Simon task. Another possible reason why we did not observe a significantly smaller ERN/Ne associated with sleep inertia might be that stronger conflict in our task induced salience of errors, and thus there was no deterioration of error detection processing.
Balkin et al. (2002) showed that the anterior cingulate cortex (ACC) was rapidly reactivated after awaking from sleep, and did not vary significantly between 5 and 20 min after awaking in contrast to increased activity of the prefrontal cortex during the same period. The ERN/Ne is thought to be generated in the ACC (e.g. Holroyd et al., 1998; Kiehl et al., 2000); therefore, no effect of sleep inertia on ERN/Ne in this study might be due to the quick reactivation of the ACC after sleep offset. In contrast, the sleepiness resulting from sleep deprivation is surmised to be associated with deactivation of broad brain areas including the ACC (Thomas et al., 2000). This speculation is supported by the results of Tsai et al. (2005) and Hsieh et al. (2007), which showed reduced ERN/Ne after one night sleep deprivation. Therefore, the discrepancy among previous findings about the effect of sleepiness on ERN/Ne might also be ascribed to different neurocognitive function during sleep deprivation and sleep inertia.
Recently, some researchers recommended that a nap should be taken during night shifts to maintain the vigilance level for rotation shift workers (e.g. Purnell et al., 2002; Sallinen et al., 1998; Smith et al., 2007; Takeyama et al., 2005). These workers must arise near the trough period of core body temperature and they very likely have some sleep dept before their nighttime nap. Therefore, their sleep inertia after a nap might be more severe than the participants in this study were exposed to. Considering our finding that Pe amplitude was clearly reduced even without any previous sleep deprivation prior to the nap and time of day they were awakened was not the trough period of body temperature, this decline in error processing may be more severe immediately after a nap during the evening hours, or during a night shift.
It should be noted that there were some limitations in this research. First, the small sample size of this study might increase the possibility of type 2 error. To address this problem, we calculated the effect size for each analysis and, with the exception of two behavioral measures, they were generally small (pη2 < 0.10). Therefore, these results may have been affected by the small sample size. However, the analyzes of primary interest showed clear results. For example, the effect size of the significant interaction associated with the Pe (pη2 = 0.63) was far larger than that for the ERN/Ne (pη2 < 0.001). Therefore, it seems unlikely that even a larger sample would yield a significant difference in the ERN/Ne amplitude based on this effect size (pη2 < 0.001).
Second, to keep the duration of nap constant, we could not control the sleep stage that participants were awakened from. It might be noteworthy that the amplitude of the Pe declined during sleep inertia even though not all participants were awakened from SWS. However, the effect of sleep stage from which participants were awakened on performance monitoring is unclear. Further research with sample sizes large enough to compare Pe amplitude across sleep stage from which participants are awakened or with the procedure fixed such that participants are all awakened from a predetermined sleep stage would help address this issue.
Third, to ensure an adequate number of epochs were used for each ERP average, the data of four blocks in each session were combined. This procedure prevented us from exploring the time course of sleep inertia. Further research examining this issue might provide meaningful insights into the duration and characteristics of sleep inertia and the appropriate recovery required before workers can safely resume their duties after a nap.
This study suggests that participants tend to underestimate the significance of their own erroneous responses during sleep inertia. Therefore, it may be very important for individuals relying on a nap to preserve vigilance and attention, especially during a night shift, to employ sufficient countermeasures to minimize any negative effects of sleep inertia after a nap. These may include: shortening nap duration to <20 min (e.g. Fushimi and Hayashi, 2008; Hayashi et al., 1999a,b), using caffeine (e.g. Hayashi et al., 2003b; Van Dongen et al., 2001) or scheduling a wake-up period after longer naps. These safeguards are especially important to promote because in this study participants did not report any sleepiness after a 1-h nap. Therefore, any effect of sleep inertia on performance monitoring may be insidious and discounted by workers.
This research was supported by KAKENHI [Grant-in-Aid for Young Scientists (B), no.18730472, 20730401] from the Japan Society for the Promotion of Science to the S. A., and by the Grant-in-Aid for Scientific Research (C) 21530774 to H. M.