How Short and Full Sleep Loss Affect Thinking Skills at Different Times and in Different People

Thirty-six healthy young men and women (see Table S1 for demographic information) participated in this cross-over design (Figure 1A). Average polysomnographically assessed total sleep time (TST) per 24 h was 8.56±0.06 h in the Control condition and only 5.75±0.06 h in the Sleep Restriction (SR) condition, in which participants were given a 6-h sleep opportunity, i.e. under repeated PSD (SR vs. Control: F_1,72.2 = 2075.68, P<0.0001; Figure 1B). During the 12-h recovery sleep episode following acute TSD, TST was significantly longer in the SR than in the Control condition (11.19±0.16 h vs.10.86±0.16 h, F_1,28 = 5.23, P = 0.03; Figure 1B). Thus, seven nights of sleep restriction induced a sleep deficit that was carried over the period of acute TSD.

Participants repeatedly performed a test battery throughout the protocol (Figure 1A; see Materials and Methods for details of the test battery).

Subjective Alertness, i.e. the reverse of the participants’ scores on the Karolinska Sleepiness Scale [31], was stable during the Control condition, but declined steadily during the sleep restriction days (D1 to D6; see Figure 1C). In the subsequent TSD period, Subjective Alertness declined markedly from the first day (TD1), through the following biological night (TN1), with a limited recovery the next day (TD2).

Sustained Attention was assessed with the Psychomotor Vigilance Task (PVT) [32] in which participants wait an unpredictable amount of time for a single discrete stimulus before making a simple, well-practised response. Response speed of the 10% slowest responses deteriorated during PSD and TSD (Figure 1D; see Figure S1A and Supporting Information for the results of lapses). Sustained Attention was further evaluated with the Sustained Attention to Response Task (SART) [33] in which a simple, well-practised response is made to the vast majority of rapidly presented stimuli, and withheld when a target stimulus is presented. Performance on the SART was quantified taking both errors of omission and commission into account by computing an accuracy measure, A’ (refer to the statistical methods section for details), which deteriorated during PSD and TSD (Figure 1D).

We assessed Working Memory and Executive Functions with verbal n-back tasks in which participants repeatedly compare the current stimulus with the one presented 1 to 3 items before. As task difficulty increases from 1- to 3-back, so does the load on the three components of Executive Functions, i.e. updating of maintained information, task switching (e.g. between updating and memory comparison), and response selection [16]. Performance which was also quantified by A’ varied significantly across the three n-back tasks with poorest performance observed on the 3-back (main effect of Level on A’: F_2,68 = 26.71, P<0.0001 and for each pair-wise comparison, P<0.0001; on a measure of response tendency, bias [B”_D]: F_2,68 = 29.92, P<0.0001 and for each pair-wise comparison, P<0.05). Although the negative effect of TSD on n-back performance was substantial, especially during TN1, and a smaller effect of PSD was observed, n-back tasks with a higher executive load did not appear to be more affected by sleep deprivation than those with a lower executive load (Figure 1E). On the other hand, B”_D was more sensitive to the effects of PSD. After two or four nights of sleep restriction, participants became more conservative in their responses, i.e. increased their tendency to claim that ‘matches’ were not occurring (see Figure S1B).

For the statistical evaluation of these effects and for the comparison of the size of the effects across tasks, we first analysed all segments of the protocol combined (i.e. baseline, D1–D6, TD1, TN1, and TD2; see Table S2 for statistical results). We computed the implied effect size (f²; refer to the statistical methods section for details) and found that the effect of Condition, i.e. Sleep Restriction vs. Control, was greater for Subjective Alertness and Sustained Attention measures than for Working Memory (Figure S2). For the latter tasks, the effects of Condition were small and independent of the level of executive demand (Condition x Level interaction on A′: F_2,68 = 0.38, P = 0.68; B″_D: F_2,68 = 0.47, P = 0.63).

We next determined whether repeated PSD and acute TSD had differential effects on performance by contrasting effects in different segments of the protocol. The effect of repeated PSD was estimated by comparing performance after five and six nights (D5 and D6) of restricted sleep opportunities to performance during D5 and D6 in the Control condition. We found a significant effect of repeated PSD on Subjective Alertness, Sustained Attention, and all measures of Working Memory performance except for B″_D in the verbal 3-back task (Table S3). The effect size of repeated PSD varied considerably across the various cognitive domains. By conventional metrics [34], the effect size was large for Subjective Alertness, medium for the Sustained Attention measures, and small for Working Memory (Figure 2A). For the Working Memory tasks, the effect of repeated PSD was independent of the level of executive load (Level × Condition interaction on A′: F_2,67 = 0.14, P = 0.87; B″_D: F_2,67 = 0.58, P = 0.56).

When we contrasted all the 52 performance measures of the seven cognitive domains between the Control and the SR conditions, we found a near linear increase in the effect of repeated PSD from D1 to D6 (). Figure S3

To further examine the differential effects of repeated PSD on Sustained Attention and Executive Functions, we contrasted the change of performance at the end of the sleep history manipulation period relative to the baseline day (i.e. D6/baseline) in the SR and the Control conditions. After six nights of restricted sleep opportunities, PVT performance was at a 79.86±1.04% (mean speed of the slowest 10% responses ± SEM) level relative to baseline, while at the end of the Control period, performance was at a 93.38±1.04% level, and these changes in performance across the sleep history manipulation period differed between the two conditions (F_1,33.8 = 11.49, p = 0.002). In contrast, verbal 3-back performance (A′) at the end of the sleep history manipulation period was at a 99.36±1.01% level in the SR condition and 100.15±1.01% level in the Control condition relative to baseline, and no significant main effect of Condition was found (F_1,34.1 = 0.19, p = 0.66). These results emphasized that repeated PSD impaired Sustained Attention, while its influence on Executive Functions was minimal.

The effect of acute TSD was estimated by comparing performance on TD1 to performance on TD2 of the TSD segment. These performance measures were obtained at the same circadian phase and only differed with respect to the duration of the wake episode preceding the performance assessments. TSD impaired all performance measures (Table S3), but the detrimental effects were more prominent for Subjective Alertness and Sustained Attention measures than for Working Memory (Figure 2B). The effect of TSD on Working Memory was independent of the level of executive load (Level × Condition interaction on A′: F_2,175 = 1.12, P = 0.33; B″_D: F_2,175 = 2.95, P = 0.05).

To further investigate the differential effects of acute TSD on Sustained Attention and Executive Functions, we contrasted the change of performance on the second relative to the first day of the TSD period (i.e. TD2/TD1) in the SR and the Control conditions. Relative to the first day of TSD, PVT speed on the second day was reduced by about 50% in both the SR and the Control conditions (mean ± SEM: 46.28±1.08% vs. 49.46±1.08%, F_1,32.7 = 0.48, p = 0.49), while verbal 3-back performance was maintained at an approximately 90% level (A’ in SR vs. Control: 90.64±1.03% vs. 91.81±1.03%, F_1,34.1 = 0.14, p = 0.71). These findings further indicated the greater detrimental effects of acute TSD on Sustained Attention than on Executive Functions.

We also evaluated the effects of TSD on performance during the circadian night by contrasting performance on TD1 and TN1. As in the previous comparisons, the effects of TSD were larger on Subjective Alertness and Sustained Attention (particularly reflected in PVT speed) than on Working Memory performance (Figure 2C).

This indicates the robustness of these differential effects of sleep deprivation on various cognitive domains.

The effects of repeated PSD and acute TSD were computed for other measures of these tasks, e.g. lapses of attention and speed on the 10% fastest response in the PVT, errors of omission and commission of the SART, as well as tasks including measures of Motor and Temporal Control and subjective assessments of Workload (see). Even when all these measures were considered, Sustained Attention and subjective measures, including those of Workload, remained most affected by the sleep manipulations and particularly so by acute TSD. Figure S4A and B

The three PER3 genotypes obtained similar amounts of sleep (main effect of Genotype on TST: F_2,33.9 = 0.89, P = 0.42) and their TST was not differentially affected by the two conditions (Condition × Genotype: F_2,76.1 = 0.74, P = 0.48).

The time course for selected performance measures throughout the protocols was assessed separately for the three genotypes. When the entire protocol was considered, there was no significant effect of Genotype on Subjective Alertness, but the interaction between Genotype and Condition was significant (Figure 3A and B; Table S4) such that the negative effect of sleep restriction on Subjective Alertness was greatest in the PER3^5/5 participants. The effect of Genotype or Genotype × Condition interaction was not significant for the Sustained Attention measures (Figure 3C; see Figure S5A for the time course and Table S4 for the statistics). When all segments of the protocol were considered, no significant effect of Genotype was observed for any of the Working Memory measures, and the interaction between Genotype and Condition was significant in just one circumstance (Table S4). Only for A’ in the verbal n-back task with the highest executive load, i.e. verbal 3-back, did we observe a significant interaction (F_2,155 = 3.42, P = 0.04; Figure 3A and D), indicating that sleep restriction affected 3-back accuracy differently across the PER3 genotypes. Specifically, the performance of the PER3^4/4 homozygotes was not affected by sleep history during either the sleep restriction days or subsequent TSD, while in the PER3^4/5 heterozygotes, negative effects of sleep restriction were observed during the TSD period only. In the PER3^5/5 homozygotes, a decrement in verbal 3-back performance was quite consistently found after the first night of 6-h sleep opportunity, and was also observed during subsequent TSD (see Figure S5B for a similar time course of B”_D). These differential responses of the PER3 genotypes to sleep loss were not found in Working Memory tasks with a smaller executive load, i.e. 2-back, or in relatively undemanding Working Memory, i.e. 1-back (Figure 3D).

Comparison of the effect size of the interaction between Genotype and Condition across these tasks indicates that when all segments of the protocol are considered, Subjective Alertness and verbal 3-back performance during sleep loss were most influenced by Genotype (Figure S6). The PER3 genotypes did not differ in their responses to sleep deprivation for Sustained Attention or Working Memory tasks with small load on Executive Functions.

Analyses conducted separately for the effects of repeated PSD and acute TSD revealed that whereas the Genotype effect on Subjective Alertness was more pronounced for PSD (Figure 4A), the Genotype effect on Executive Functions was almost exclusively related to the TSD effect (Figure 4B). Thus, the f² of the Genotype × PSD interaction on KSS score was 5.5 times larger than that of the Genotype × TSD interaction. The f² of the Genotype × TSD interaction on verbal 2-back performance was 18.2 times larger than that of the Genotype × PSD interaction. For speed measures on the PVT, effect sizes of Genotype were relatively small for both PSD and TSD. Thus, largest effects of PER3 genotype during acute TSD concern different cognitive domains than during repeated PSD.

Please note that even in the PER3^5/5 participants, the effect of TSD was greater on Sustained Attention (the speed of the slowest 10% responses in the PVT) than on Working Memory (Figure S7).

Circadian rhythmicity itself was affected by PSD. After repeated PSD, the circadian rhythm of melatonin was delayed by 45 minutes (mean ± standard error of dim light melatonin onset [DLMO]: 00∶13±00∶15 vs. 23∶28±00∶15; main effect of Condition: F_1,30 = 28.15, P<0.0001; Figure S8; see Table S5 for the effects of Condition and Genotype on other circadian phase markers). For all subsequent analyses of the effect of Circadian Phase, the behavioural data during TSD were aligned to the melatonin rhythm.

Prominent circadian variation was observed for Subjective Alertness, Sustained Attention, and Working Memory during the TSD period (Table S6). For all of these measures, the minimum was observed in the morning hours, i.e. DLMO +8 h (Figure 5A–C). Following repeated PSD, all performance measures were lower throughout the TSD period (Table S6 and Figure 5A–C). However, the effect of repeated PSD on the time course of performance during TSD was not constant (Figure 5A–C). For most of the measures, the effect was greatest in the initial part of the TSD period, then became smaller, with a minimum in the evening hours close to the DLMO, and increased again during the circadian night with a maximum at around DLMO+4 h. On the second day of TSD, effect of prior PSD was, in most cases, smaller (Figure S9). Indeed, we found a significant Condition × Circadian Phase interaction for Subjective Alertness and PVT measures, though not for SART or n-back measures (Table S6).

For the Working Memory performance measures, pairwise comparisons of performance per circadian bin suggest that significant effects of prior PSD were mainly observed in the initial part of the TSD period and during the biological night at DLMO +4 h (Figure 5C).

Averaging the effects sizes of prior PSD across these cognitive domains but separately per circadian phase bin revealed the general pattern of largest effects in the morning and afternoon during the initial hours of TSD, followed by a decline of the effects with a minimum in the evening hours at around the DLMO, followed by a sudden subsequent increase during the biological night (Figure 5D; effect of Circadian Phase: F_8,40 = 6.33, P<0.0001). This circadian modulation of the effects of sleep history on performance decrements during TSD was confirmed by analysing the effect sizes of all measures of the performance battery including Temporal and Motor Control and subjective assessments of Workload (see Figure S10A; effect of Circadian Phase: F_8,408 = 28.96, P<0.0001).

PER3 genotype modulated the interaction between Circadian Rhythmicity and Sleep History but to a different extent for Executive Functions than Subjective Alertness, Sustained Attention, and Working Memory with a low executive load. In the PER3^4/4 homozygotes, verbal 3-back performance during sustained wakefulness was not affected by prior PSD at any circadian phase. By contrast, in the PER3^4/5 heterozygotes and the PER3^5/5 homozygotes, verbal 3-back performance was poorer following PSD especially at DLMO+4 h, which on average corresponded to 02∶00 to 06∶00 (Figure 6A). This circadian modulation of the effect of Genotype × Sleep History interaction was confirmed by comparing the effect size of the Genotype × Condition interaction on performance during TSD at various circadian phases for Subjective Alertness, Sustained Attention, and the verbal n-back tasks. Although for most of the performance measures, effect sizes were in general larger at the beginning of the TSD period during the circadian night, and in the post-TSD morning than in the post-TSD afternoon, the largest f² was observed for the verbal 3-back performance at DLMO+4 h, i.e. between 02∶00 and 06∶00 (Figure 6B). A general pattern of largest effects of Genotype in the morning during the initial hours of the TSD, followed by a decline of the effects throughout the biological day and a sudden subsequent increase during the biological night was revealed by averaging the effects sizes of Genotype across these cognitive domains but separately per circadian phase bin (Figure 6C; effect of Circadian Phase: F_8,40 = 2.43, P = 0.03). The standard errors of the effect sizes co-varied with the average effect size, indicating that not only the effect sizes vary with circadian phase but also the extent to which different cognitive domains are differentially affected. This was further confirmed by analysing the effect sizes of all measures of the performance battery (see Figure S10B; effect of Circadian Phase: F_8,408 = 2.65, P<0.01), such that largest effects of Genotype and largest divergence between tasks were observed in the morning hours, with smaller effects observed in the evening and afternoon.

The longer TST during recovery sleep after Sleep Restriction relative to the Control conditions (Figure 1B) suggests that limiting the sleep opportunity to 6 h per 24 h was successful in creating a chronic sleep debt that was carried over the period of TSD. The relatively small difference in TST (approximately 20 min) during the recovery sleep episodes probably reflects a ceiling effect since TST was close to 11 hours during the 12 h sleep opportunity. In the Control condition, during which performance did not deteriorate, participants on average slept 8.56 h. This is close to a previous estimate of the sleep duration required to maintain performance [8], [9] and only slightly less than the 8.9 h which is the estimated maximal capacity for sleep in young adults when given 16 h of sleep opportunity per 24 h [35].

The data reported above provide several indications for the validity of our performance measures. The magnitude of changes in Subjective Alertness observed in the current study, are comparable to previous reports [8], [9]. Our measures of Sustained Attention behaved as expected: the speed of the 10% slowest responses on the PVT was much more affected by repeated PSD and acute TSD than the 10% fastest responses. This is in accordance with previous behavioural studies [36], [37] and provides evidence for differential brain correlates of the slowest and fastest responses as assessed in fMRI studies [38], [39]. Consistent with previous reports on the effects of sleep deprivation on the SART, which involves Sustained Attention and Response Inhibition [40], both errors of omission and commission were affected by sleep deprivation, but in our data, the accuracy measure, A’, was shown to be more sensitive than either of the error measures (Figure S4). Performance on the Working Memory tasks with varying executive load displayed characteristics in accordance with the intended targeted processes. The progressive reduction of performance from 1- to 2- to 3-backs is consistent with existing findings [27] and was observed throughout the protocol. Moreover, as A’ was above chance level, we are confident that the observed decrease in accuracy and increase in tendency towards more conservative responses were associated with task difficulty and that participants continued to engage genuinely with the tasks even after repeated administration.

In the present study, we compared the effects of sleep deprivation across tasks carefully selected to recruit different cognitive domains. These tasks have been widely used in cognitive research, in both laboratory and neuroimaging studies, as well as in sleep research. We used multiple tasks, with known redundancies and partial functional overlaps (e.g. 1-back verbal, spatial, visual, integrated spatial and visual) in order to counter the inherent difficulty with cognitive tasks that no task is ‘process pure’. That is, even apparently simple tasks may depend upon multiple simple processes. These simpler processes are themselves more or less influential in performance as tasks are, wittingly or otherwise, performed differently by participants on different occasions (e.g. context, time-pressure, practice, speed-accuracy trade-offs, feedback, reward, etc.). By using multiple tasks that had been extensively practiced, and almost all of which are externally paced, with minimal direct knowledge of results, we believe we can minimise many of these unwanted sources of performance variability. Our multi-task approach provides an alternative to other approaches in which the effects of sleep deprivation are investigated on the cognitive components of a single task, the interaction between which is prey to all of the influences identified above [22].

Computation of effect sizes [34], [41] and the simultaneous and repeated assessment of performance in many cognitive domains allowed for a unequivocal demonstration that the effects of repeated PSD and acute TSD on Subjective Alertness and Sustained Attention are larger than effects on Working Memory/Executive Functions. In fact, little evidence for a particular sensitivity of the executive component of the n-back tasks to the effects of sleep loss was observed in the present study. Importantly, no interaction between executive load and either PSD or TSD was observed for A′ or B″_D. The fact that throughout the present study, performance was reliably affected by task difficulty, allows us to be more conclusive regarding the lack of a relationship between task difficulty and the negative effects of sleep deprivation, since in previous studies, simpler and more difficult tasks were impaired by sleep deprivation to similar extents [42], [43], [44].

Meta-analyses of studies on the effects of acute TSD on accuracy and speed measures of performance across a wide range of cognitive domains indicated that Sustained Attention was particularly affected, although only few studies included in the meta-analyses used n-back tasks [20] and none assessed the effect of repeated PSD on Executive Functions as assessed by the n-back tasks. In our study, effects of sleep deprivation on Sustained Attention were not only greater than effects on Working Memory, but also greater than effects on measures of Temporal and Motor Control (Figure S4). The differential sensitivity of performance across cognitive domains was observed during the circadian night and day and following repeated PSD and acute TSD. In addition, in the current data set, the effects of TSD were larger than the effects of PSD, but the relative sensitivity of the various tasks appeared similar for the two interventions.

We used effects sizes, which are widely used in the psychological literature and also in meta-analyses, as our primary tool to compare the effects of sleep deprivation across tasks. The validity of this choice was confirmed by computing relative changes in performance for some of the tasks and these measures also showed that Sustained Attention task was more affected than Working Memory task even when implemented with a high executive load. In the current protocol, we also directly compared the effects of PSD and TSD and across many different tasks, thereby providing yardsticks both within and across tasks. Effect sizes, although useful in comparing across tasks, interventions, and studies, do, of course, not inform about the ‘clinical’ significance of the effect. This would require comparing the effects of sleep loss on tasks to the effects observed in specific patient populations, or using tasks that have in a quantitative way been correlated with performance in the real world.

In our protocol, sleep restriction was accomplished by a symmetrical change in bedtime and wake time because we wanted to keep the centre of light exposure unchanged and thereby minimise circadian phase shifts. Nevertheless, repeated PSD led to a delay of the melatonin rhythm. This finding provides further evidence [45] for a feedback of the sleep homeostat onto the circadian timing system and emphasizes the necessity to assess circadian phase. In contrast to previous studies [22], [46], performance data were aligned with the melatonin rhythm, which is particularly important when group averages are computed across individuals with different circadian phases. These analyses demonstrated that the effects of sleep history on cognition were modulated by circadian phase. During acute TSD, all cognitive performance measures displayed a near stable performance during the first day, a steep decline during the biological night, and some recovery on the second day [10], [47], suggesting that one circadian signal modulates all these performance measures.

The effects of sleep history were greatest in the initial part of the acute TSD, became progressively smaller and then increased again during the biological night. In our protocol, sleep restriction was accomplished by a symmetrical change in bed time and wake time; hence, compared to the Control condition, in the SR condition, wake time occurred at an earlier clock time (2 h) and at any given clock time, participants had been awake for a longer duration. This can, however, not be the only explanation for the time course of the effects of sleep history because even we compared a Sleep Restriction observation at for example 12∶00 to a Control observation at 16∶00, Alertness in the SR condition was still poorer (see Figure 5A). The time course of the effect of sleep history on performance indicates that a circadian signal can modulate the effects of sleep history on Subjective Alertness, Sustained Attention, and Executive Functions, in accordance with similar observations for Sustained Attention [12]. During the wake maintenance zone when the circadian wake promoting signal is at its crest [48], the brain is relatively protected from the negative effects of sleep deprivation on waking performance. By contrast, in the morning hours, the brain is very susceptible, and all cognitive domains are affected. This circadian modulation of sleep history effects appears strongest for Sustained Attention measures, as reflected in a significant interaction between Condition and Circadian Phase for these measures and in accordance with a previous report [12] in which only Sustained Attention was assessed. The neurochemical basis of this circadian protection of cognition remains to be elucidated, although some suggestions have been made [49].

At baseline, no differences in performance between the PER3 genotypes were observed. Performance deteriorated following manipulation of sleep homeostasis by both repeated PSD and acute TSD in all three genotypes, but more so in the PER3^5/5 homozygotes and in a cognitive-domain-dependent manner [29]. Please note that these differences between the genotypes cannot be explained by differences in sleep timing between the genotypes (see Table S1) because all the events in the laboratory were scheduled relative to habitual sleep times. The more rapid reduction in daytime Subjective Alertness in response to repeated PSD in PER3^5/5 homozygotes is a new observation [50], whereas the absence of a Sleep Loss × Genotype interaction on Sustained Attention after both repeated PSD and acute TSD is consistent with previous studies [29], [50], [51]. The exquisite dependency of the expression of the Genotype effect on sleep pressure and circadian phase is in accordance with our previous behavioural [29] and fMRI studies [30] and demonstrates the importance of considering circadian phase and homeostatic sleep pressure in the study of individual differences [24], [25]. In fact, including all performance measures in the analyses of the interaction between homeostatic sleep pressure, circadian phase, and genotype demonstrated that the greatest divergence between individuals and cognitive domains was observed in the early morning hours after sleep deprivation. Thus, sleep deprivation in combination with circadian phase misalignment is a good paradigm to identify individual differences and task specificity in the susceptibility to performance decline.

Impairments in performance induced by sleep deprivation have been interpreted within various frameworks, such as the ‘neuropsychological hypothesis’, the ‘vigilance hypothesis’, and the ‘controlled attention hypothesis’ [20]. According to the ‘neuropsychological hypothesis’, sleep deprivation leads to a temporary ‘functional lesion’, particularly in the frontal and prefrontal areas [18], and behavioural deficits in those cognitive domains that rely on these brain areas, such as Executive Functions. Our data clearly indicate that Executive Functions are not more prone to the negative effects of sleep deprivation than Sustained Attention. Furthermore, performance in the n-back tasks that are more difficult and depend heavily on prefrontal mediation are not more affected than simpler versions [52].

The ‘vigilance hypothesis’ states that Sustained Attention is much affected by sleep deprivation and that Sustained Attention is very important to higher aspects of cognition [20]. Although our data show the sensitivity of Sustained Attention tasks to sleep deprivation, they also demonstrate that at the same time, higher aspects of cognition are not very much affected.

By contrast, our data may seem to favour, to some extent, the ‘controlled attention hypothesis’ [53] in which task characteristics related to how engaging they are determine the extent to which they are impaired by sleep deprivation, with the simpler, more monotonous, and less engaging tasks being more affected because in these tasks, ‘top-down’ control which is required for optimal performance is depleted during sleep deprivation. We do not have an independent measure of task engagement, but would note that the two Sustained Attention tasks used in the present study (PVT and SART) are simpler and more monotonous than those tasks assessing Working Memory and Executive Functions (the n-back tasks). In the PVT, participants are exposed to the same stimulus (onset of a time counter) for 10 minutes and need to make a simple response at random intervals ranging between 2 and 10 seconds. Compared to the PVT, the SART is less monotonous because participants are required to decide whether the stimulus displayed is a target or non-target and either withhold or make a response every 1,150 ms for about two minutes. Even more complex and challenging are the verbal n-back tasks in which every 2,500 ms, participants need to update their memory of 1–3 stimuli, compare the current stimulus with the most remote one in their memory, and decide whether it is a match or not, while switching between all of these task components. Due to the differences in the characteristics of these tasks, we postulate that the Working Memory/Executive Function tasks are the most complex, challenging, and engaging tasks, and hence, less affected by sleep deprivation. In fact, some studies have shown that adding an additional item manipulation requirement to a Working Memory task, i.e. increasing the complexity and possibly the degree of participants’ engagement to the task, renders the task less susceptible to the negative effects of sleep deprivation [54].

The observed small deficits after sleep deprivation in more demanding or engaging tasks, the self-reported effort to perform these tasks, and fMRI data [52] all indicate, however, that the top-down executive control of attention required to engage with and perform these task is ultimately affected by sleep deprivation and, in part, forms the basis of inter-individual differences. This theoretical account and our data have a number of practical implications. Most obviously, for young adults, a 6-h sleep opportunity is not sufficient to maintain brain functions. Furthermore, simple measures of Sustained Attention, although sensitive to sleep deprivation, may not predict how well individuals perform on other tasks [55]. Finally, circadian phase is a powerful determinant of the expression of differences in performance between individuals and tasks and identifying the mechanisms by which circadian rhythmicity accomplishes protection against the detrimental effects of sleep loss may provide new tools to improve performance or prevent the deterioration thereof.

Our data show that the extent to which acute TSD leads to deterioration in performance depends on the task domain, prior sleep debt, circadian phase at which performance is assessed, and genetically-determined subject characteristics. Careful consideration of the task characteristics in real life working conditions during night shifts and sustained operations may reduce the risks of performance failures.

Our paradigm clearly demonstrates an interaction between the circadian and the homeostatic processes in regulating cognitive performance and implies that Sustained Attention and subjective measures of Alertness and Workload are most affected by both chronic sleep restriction and acute total sleep loss. The contribution of the circadian process was assessed during the acute TSD period. The time course of performance, with a minimum early in the morning and some indications of recovery during the next day clearly indicates the contribution of circadian processes. However, during TSD, both time awake and circadian phase change simultaneously and this paradigm does not allow the examination of the individual contribution of circadian rhythmicity and the sleep homeostat to cognitive performance. The separate contribution of circadian rhythmicity and sleep homeostasis can be assessed in forced desynchrony protocols [10], in which sleep and wake episodes are systematically shifted across the circadian cycle. In future research, multiple cognitive domains should be examined in a forced desynchrony protocol in order to further establish whether the homeostatic and the circadian influence on performance varies across cognitive domains.

In addition, the differences in effect sizes between tasks may be related to the cognitive domain targeted by a specific task, but may, of course, also be related to other aspects of the task such as its duration. Future studies in which time on task effects are analysed across cognitive domains may be of interest.

The present study consisted of three phases: (1) telephone screening and screening visit, (2) pre-laboratory field study, and (3) laboratory study. Participants were recruited through flyers, emails, and newspaper and radio advertisements. Out of the 358 individuals who were successfully genotyped after the screening visit, 165 (46.1%) were PER3^4/4 homozygotes, 159 (44.4%) were PER3^4/5 heterozygotes, and 34 (9.5%) were PER3^5/5 homozygotes. The relative prevalence of the three genotypes in this sample is in accordance with previous reports [28], [56], [57]. Thirty-six healthy volunteers were selected and participated in the laboratory study (18 males; mean ± SD of age = 27.6±4.0 years). They were in general good health by medical history, physical examination, and standard biochemistry and haematology tests. They did not suffer from sleep disorders based on self-report questionnaires (Pittsburgh Sleep Quality Index≤5) and a clinical polysomnographical recording conducted during the first night in the laboratory. They reported consuming≤300 mg of caffeine per day and ≤14 units of alcohol per week, and were not smokers or shift workers. They had not travelled across more than one time zone two months before the laboratory phase and had not donated blood six months before. Female volunteers were not pregnant.

This sample consisted of 12 PER3^4/4, 10 PER3^4/5, and 14 PER3^5/5 who were matched for age, gender, body mass index (BMI), and ethnicity at the group level (Table S1). The three genotypes also did not differ in their morningness-eveningness preference [58], self-reported sleep quality [59], level of subjective sleepiness [31], [60], personality [61], [62], mental and physical health [63], [64], eating behaviour [65], intelligence [66], [67], and mood state [68] (Table S1). However, genotype-dependent differences were found in actigraphically assessed sleep-wake timing with PER3^5/5 individuals going to bed and waking up earlier than PER3^4/4s (Table S1), which is in accordance with a recent large-scale epidemiological study [57]. Three participants (one PER3^4/5 and two PER3^5/5 individuals) withdrew from the study for personal reasons at various points during the laboratory phase.

Note that this sample was independent of the sample used in our previous behaviour study [29], [51] and also independent of the sample used in our previous fMRI study [30].

This protocol was approved by the Institutional Review Board of the Air Force Research Laboratory and received a favourable opinion from the University of Surrey Ethics Committee. It was conducted in accordance with the principles of the Declaration of Helsinki. All the participants provided written informed consent after receiving a detailed explanation of the aims and procedures of the study.

Statistical analyses were performed with SAS 9.1 (SAS Institute, Cary, NC). We used a general linear mixed model with PROC MIXED to determine the effects of Genotype (PER3^4/4/PER3^4/5/PER3^5/5), Condition (Control/SR), Session (first/second laboratory session), and the Genotype × Condition interaction, as fixed model effects, on DLMO and other circadian phase markers, with the Subject effect as a random factor. For the performance data, we also examined the effects of Day (from baseline to the second day of TSD) as a repeated effect with a spatial power variance-covariance matrix being specified. For the performance data during the TSD period, we first aligned the data to the melatonin rhythm, and instead of the Day effect, we examined the Circadian effect (from DLMO −12 h to DLMO +20 h; 4-h bins) as a repeated effect with a first-order autoregressive variance-covariance matrix being specified. Differences of least square means were used to determine significant differences among the PER3 genotypes and between the two conditions at P<0.05.

Sleep data during the habituation night were not included in the analyses. Performance data collected on the arrival and habituation days were also not included in the analyses to minimize the effect of learning and practice on the results. Due to technical problems and drop-outs, 2.8% of the performance data, 8.6% of the PSG records, and 6.4% of the blood samples could not be included in the analyses.

Effect sizes were indicated by Cohen’s f²[34], [41]:where u and v are respectively the numerator and denominator degrees of freedom of the F statistic used to determine the corresponding main or interaction effect in the general linear mixed model analysis.

In addition to using effect sizes to quantify the effects of repeated PSD on Sustained Attention and Executive Functions, we also compared the magnitude of change in performance in the PVT and the verbal 3-back task from the baseline day to the end of the partial sleep deprivation period (expressed as the performance on D6 in the SR condition divided by the performance on the baseline day) with the change in performance across the same days in the Control condition. For the effect of acute TSD, we did a similar comparison between the change in performance from the first to the second day of the TSD period between the two conditions (expressed as performance on TD2/performance on TD1).