1 INTRODUCTION

Human manipulation of the timing and duration of sleep is pervasive. One of the more common sleep–wake patterns in industrialized societies involves sleep restriction on weekdays followed by sleep extension on one or two weekend days off to recover (Basner et al., 2007; Holding et al., 2020; Petersen et al., 2017). The foundational two-process model of sleep regulation (Borbély, 1982; Daan et al., 1984) posits that the effects of sleep and sleep deprivation on waking neurobehavioural functioning are governed by a homeostatic process modulated by a circadian process (Borbély & Achermann, 1999). The involvement of a homeostatic process implies the existence of stable states of equilibrium (Aeschbach et al., 1996; Skorucak et al., 2018), consistent with the long-standing idea that people could cut back on their sleep without adverse consequences for neurobehavioural performance (Horne, 1985).

However, this idea has been refuted by mounting evidence that exposure to consecutive days of sleep restriction causes a substantial, steady build-up of neurobehavioural performance impairment across days of sleep restriction (Belenky et al., 2003; Cohen et al., 2010; Dinges et al., 1997; McHill et al., 2018; Van Dongen et al., 2003). This accumulation of impairment has been shown to be sleep dose-dependent, where the more truncated the sleep duration (e.g. 4 versus 6 hr of sleep), the greater the magnitude of impairment (Belenky et al., 2003; Van Dongen et al., 2003). Intermittent days of recovery sleep have been shown to lessen the accumulated impairment, but neurobehavioural performance is only partially restored (Lo et al., 2022; Smith et al., 2021), with full recuperation of neurobehavioural functioning requiring many consecutive days of recovery sleep (Banks et al., 2010; Belenky et al., 2003). These findings led to a paradigm shift in our understanding of the impact of sleep deprivation on neurobehavioural functioning (Banks & Dinges, 2007), and gave rise to the premise that prior sleep loss increases sensitivity to further sleep loss (Grant & Van Dongen, 2013). However, it remains unclear what the trajectory of neurobehavioural impairment across days of sleep restriction would be, for instance, in the face of intermittent days of more severe sleep restriction, as may be experienced by medical residents and others in 24/7 operations. Notably, the general properties of the dynamic regulation of the build-up and dissipation of neurobehavioural impairment across periods of sustained sleep restriction are as yet unknown.

We implemented this design in a laboratory-based dose–response study in which healthy adults were first subjected to five consecutive days of sleep restriction to 4 hr time in bed (TIB) per night. This was followed by a 1-day DRI with randomization to a nighttime sleep opportunity ranging from 12 hr down to 0 hr TIB (in steps of 2 hr)—that is, from intermittent sleep extension down to intermittent total sleep deprivation. This was then followed by another five consecutive days of sleep restriction to 4 hr TIB per night. A non-sleep-restricted group was included as a control condition. Through this design, our experiment revealed a novel facet of sleep/wake regulation with both theoretical and practical implications for sleep science.

2 METHODS

2.1 Subjects

A total of N = 70 healthy participants (31 females, 39 males; age M ± SD = 28.9 ± 6.7 years) completed the 16-day/15-night in-laboratory experiment (Figure 1). Of these, n = 62 were assigned to two, 5-day sleep restriction (SR) periods with an intervening sleep DRI day. On the DRI day, participants were randomized to one of seven doses of sleep opportunity (i.e. TIB): 0 hr (n = 5), 2 hr (n = 7), 4 hr (n = 8), 6 hr (n = 10), 8 hr (n = 11), 10 hr (n = 11) or 12 hr (n = 10). The remaining n = 8 participants were assigned to a control condition that received a 10 hr TIB sleep opportunity on all 15 study nights, which served to rule out confounds from any non-specific effects related to the laboratory study protocol. The study was approved by the Institutional Review Board of the University of Pennsylvania. Participants provided written, informed consent. They had to meet defined inclusion criteria to be deemed eligible for the study (see Supporting Information; and see Table S1 for participant demographics).

3 RESULTS

3.1 Sleep physiology

Sleep periods on selected nights were recorded with PSG and scored for markers of sleep homeostasis (Borbély & Achermann, 1999), including TST and SWE in the electroencephalogram (EEG) of non-rapid eye movement (REM) sleep. The sleep physiological responses to the repeated 5-day periods of sleep restriction separated by an acute sleep DRI are presented in Figure 2 (see also Figure S1). TST closely matched the duration of the sleep opportunity provided on each night (Figure 2a; condition by day interaction: F_29,255 = 93.32, p < 0.001). As expected, sleep efficiency (i.e. the ratio of TST to TIB) was greater on sleep restriction nights (M ± SD = 95.6% ± 5.1%) than at baseline (M ± SD = 87.1% ± 8.7%; Figure S1d).

The TST on the DRI night showed a monotonic, near-linear dose–response effect (Figure S2a; F_5,253 = 608.50, p < 0.001), where TST increased steadily with longer sleep opportunity (although the marginal increase in TST was somewhat diminished for longer TIB, with concomitant decrease in sleep efficiency; Figure S2e). Similar patterns were found for other sleep stages, particularly stage 2 and REM sleep (Figure S2b,d, respectively). In line with previous reports of saturating slow-wave sleep duration with longer sleep opportunities (Banks et al., 2010; Van Dongen et al., 2003), slow-wave sleep duration did not exhibit a monotonic dose–response function on the DRI night (Figure S2c).

The SWE expressed relative to baseline matched the general pattern found for TST (Figure 2b; condition by day interaction: F_23,192 = 6.60, p < 0.001). SWE decreased modestly from the baseline night with 10 hr TIB (B2) to the first night of sleep restriction to 4 hr TIB (SR1), yet SWE exhibited little change from the beginning (SR1) to the end (SR5) of the first sleep restriction period. On the DRI night, there was a clear dose–response effect (Figure S2f; F_5,192 = 16.06, p < 0.001), whereby SWE increased as sleep opportunity and TST increased, albeit DRI sleep opportunities longer than 8 hr TIB exhibited similar amounts of SWE. Overall, SWE in the second sleep restriction period was not different from SWE in the first sleep restriction period (F_1,146 = 0.50, p = 0.48), although SWE on the first night of the second sleep restriction period (SR6) was lower than SWE on the fifth night of the first sleep restriction period (SR5; F_1,192 = 4.42, p = 0.037).

Overall, the sleep physiological responses to repeated 5-day periods of sleep restriction with an intervening 1-day DRI schedule perturbation were consistent with established principles of sleep homeostasis (Borbély & Achermann, 1999).

3.2 Neurobehavioural functioning

The PVT showed a near-linear accumulation of neurobehavioural impairment across the first 5-day sleep restriction period (Figure 3a,b). On the DRI day, there was a clear dose–response effect (F_6,61 = 2.71, p = 0.021), with performance improving for sleep doses greater than 4 hr TIB and deteriorating further for sleep doses of 4 hr TIB or less. Yet even the longest DRI sleep opportunity of 12 hr TIB was not sufficient to restore PVT performance to baseline levels (t₆₁ = 3.06, p = 0.003).

Of primary interest were our observations for objective neurobehavioural performance during the second 5-day sleep restriction period, which showed further accumulation of PVT performance impairment (F_7,61 = 3.26, p = 0.005). This was expected for the DRI condition of 4 hr TIB, as the sleep opportunity schedule for that condition was equivalent to a period of 11 consecutive days with sleep restriction to 4 hr TIB per night in a previous study (Van Dongen et al., 2003). However, the trajectories of neurobehavioural impairment during the second 5-day sleep restriction period were characterized predominantly by a continuation of the trajectory from the first 5-day sleep restriction period in all study conditions, regardless of the DRI sleep dose. Although the rates of accumulation of neurobehavioural impairment differed from the first to the second sleep restriction period within DRI sleep dose conditions (F_7,61 = 2.88, p = 0.012), the accumulation of neurobehavioural impairment was greater overall across the first sleep restriction period than across the second sleep restriction period (Figure 3b). Unexpectedly, the accumulation rates in the second sleep restriction period displayed no significant influence of the sleep dose from the preceding DRI night (F_6,61 = 1.33, p = 0.26). The sleep dose on the preceding DRI night produced only a slight offset difference between conditions in the second sleep restriction period (F_6,61 = 2.31, p = 0.037), in what otherwise appeared to be a steady continuation of the temporal pattern of the first 5-day sleep restriction period (Figure 3b). Such temporal dynamics were not anticipated based on the premise that the amount of prior sleep loss determines a person's sensitivity to further sleep loss, and were especially counterintuitive for the smallest (i.e. 0 and 2 hr) DRI sleep doses.

4 DISCUSSION

In this in-laboratory study of repeated exposure to sustained sleep restriction, vigilant attention measured with the PVT and subjective sleepiness measured with the KSS showed a steady accumulation of neurobehavioural impairment across days of sleep restriction (Figure 3a,b). In line with earlier findings (Van Dongen et al., 2003), the accumulation of impairment across days of sleep restriction was near-linear for the PVT, while the KSS displayed a more rapid levelling off over days. The varying amount of sleep on the 1-day DRI night was associated with acute PVT and KSS changes in a dose-dependent fashion, consistent with previous findings (Banks et al., 2010). Furthermore, for the KSS, the response to acute total sleep deprivation (i.e. the 0 hr TIB sleep dose on the DRI day) was notably more pronounced than the response to sustained sleep restriction, as documented previously (Van Dongen et al., 2003). However, regardless of the sleep dose, the impact of the DRI on the PVT and the KSS was surprisingly transient. Longer DRI sleep durations did not protect individuals from further accumulation of neurobehavioural performance deficits, and shorter DRI sleep durations did not significantly augment the deficits during re-exposure to sleep restriction (Figure 3). Thus, counterintuitively, the 1-day dose–response perturbation with a sleep opportunity ranging from as little as 0 hr TIB to as much as 12 hr TIB did not substantively alter the trajectory of subsequent accumulation of neurobehavioural performance impairment. Rather, our study revealed a tendency to continue the trend of performance changes across days regardless of the intermittent schedule perturbation, identifying a memory of sleep–wake history. Such dependence of the state of a biological system on its history is referred to as hysteresis, and it appears to be a fundamental property of the regulation of waking neurobehavioural functioning.

The dynamic response to repeated 5-day periods of sleep restriction with an intervening 1-day schedule perturbation was strikingly different for sleep physiology than for waking neurobehavioural performance. Whereas physiological markers of sleep homeostasis, TST and SWE, exhibited a robust sleep homeostatic response on the DRI night, they were largely unresponsive to sustained sleep restriction (Figure 2), as has been documented previously (Banks et al., 2010; Skorucak et al., 2018; Van Dongen et al., 2003). These observations are consistent with the principles of sleep homeostasis as instantiated in the two-process model of sleep regulation (Borbély & Achermann, 1999), for which the dynamics are determined by present homeostatic state alone and not otherwise dependent on sleep–wake history—that is, they do not involve hysteresis. The divergence between sleep physiological and waking neurobehavioural dynamics suggests that the underlying neurobiological mechanisms may be different.

In our study design, the critical assessment of the dynamics of neurobehavioural performance occurs in the second period of sleep restriction (Figure 4; SR6–SR10). During this second sleep restriction period to 4 hr TIB per day, sleep physiological markers showed the expected sleep homeostatic response, congruent with quantitative predictions of the two-process model (Figure 4a,b). Yet, the waking neurobehavioural performance outcomes exhibited much different temporal dynamics. The trajectory of neurobehavioural performance in the second sleep restriction period represented a continuation of the trajectory in the first sleep restriction period, with only a slight offset reflecting the DRI sleep dose. Thus, as demonstrated clearly by objective performance on the PVT, the temporal dynamics of neurobehavioural function from the first day immediately after the DRI day through to the last day of the second sleep restriction period reflected not only the momentaneous state of sleep loss (Figure 4c), but also the continuing influence of the effects of sleep loss from the preceding days. Quantitative predictions from a previously developed biomathematical model with a dynamic process that tracks prior sleep–wake history (McCauley et al., 2021), applied to the TIB schedules of the study (Figure 4d), match the PVT observations (Figure 4e). As such, our findings demonstrate that the neurobehavioural consequences of sleep deprivation depend not only on a person's present state of sleep loss but, also, fundamentally, on the longer-term sleep–wake dynamics that led to the present state.

One explanation for these dynamics may be that chronic sleep restriction induces neural injury that cannot be adequately reversed by just 1 night with varying sleep dose (Zhao et al., 2017). In particular, alterations to the wake-promoting ascending arousal system may contribute to the persistent neurobehavioural performance deficits resulting from chronic sleep restriction. The ascending arousal system maintains wakefulness and promotes vigilant attention through a series of brain regions originating in the brainstem, including the locus coeruleus, and extending to the basal forebrain and cerebral cortex (Saper et al., 2005). In rodent studies, extended wakefulness has been shown to reduce the number of wake-promoting neurons in the locus coeruleus through oxidative stress pathways (Zhang et al., 2014). In humans, an 8 hr recovery sleep opportunity following total sleep deprivation has been found to restore metabolic function in the frontal lobe, but not in subcortical regions (Wu et al., 2006). Because the wake-promoting ascending arousal system does not necessarily regulate the need for sleep, it could be that damage to wake-promoting neurons in the locus coeruleus due to extended wakefulness plays a role in waking neurobehavioural impairment while homeostatic sleep regulation would be unaffected.

However, an explanation based on neural damage would predict a greater, longer-lasting worsening of performance impairment from the first to the second 5-day period of sleep restriction for the 0 and 2 hr DRI conditions, as the additional sleep loss in these conditions should be associated with a further increase in the amount of neural injury sustained. Because this is not what our data showed (Figure 4e), neural injury from sleep loss cannot by itself explain the observed dynamics of waking neurobehavioural impairment from sleep loss. Alternative explanations rooted in sleep function theories regarding brain energy depletion or waste accumulation (Frank & Heller, 2019) would be similarly insufficient to account for the present observations. Still, a long-term dependence on prior states is common in pharmacokinetic-pharmacodynamic models of biochemical and physiological systems (Holford & Sheiner, 1981), including neurotransmitter-receptor systems (Louizos et al., 2014). A proposed dynamic interaction between adenosine and its receptor (Basheer et al., 2007; Elmenhorst et al., 2007), which mediate sleepiness and waking neurobehavioural function (Porkka-Heiskanen et al., 1997), would provide a plausible explanation (McCauley et al., 2009; Phillips et al., 2017).

An important implication of the observed dynamics of neurobehavioural responses across periods of sleep restriction is that sleep–wake history may have an enduring influence on an individual's vulnerability to the neurobehavioural deficits due to sleep loss (Hudson et al., 2020). On the one hand, this is important for the many people whose sleep–wake–work schedules involve curtailing their sleep on a daily basis with only intermittent opportunities for recovery sleep. As has been pointed out previously (Lo et al., 2017; Lo et al., 2019; Lo et al., 2022; Smith et al., 2021; St. Hilaire et al., 2017), 1–2 nights with extended “catch-up” sleep may be almost enough to recuperate temporarily following consecutive days of sustained sleep restriction, but a latent vulnerability to sleep loss is retained. This then manifests rapidly during a subsequent period of sleep restriction (Figure 3b), and puts people at risk to a greater degree than they may recognize subjectively (Figure 3d). On the other hand, the dynamics exposed in this study are also important for those individuals whose sleep restriction routines are punctuated by days with even greater sleep restriction or no sleep at all, as is common in sustained military operations, disaster response scenarios, and hospital rotation schedules. Our observation of the time-limited impact of this intermittent additional sleep loss (Figure 3b,d) is entirely novel.

While the participants in our study received adequate nightly sleep in the week prior to the laboratory experiment (see Supporting Information), studies of sleep extension prior to exposure to chronic sleep restriction (Rupp et al., 2009) and acute total sleep deprivation (Arnal et al., 2015) have found that extended periods of sleep up to 10 hr TIB per day can protect against the subsequent neurobehavioural consequences of sleep loss. This “banking sleep” phenomenon points to the involvement of a dynamic process tracking long-term sleep–wake history, and can be understood through the lens of hysteresis. Individuals differ considerably in their natural vulnerability to neurobehavioural impairment due to sleep loss (Rupp et al., 2012; Van Dongen, Baynard, et al., 2004), and this inter-individual variability is partially explained by heritable traits (Kuna et al., 2012; Satterfield et al., 2019). However, based on the present findings and their implications for the dynamic regulation of neurobehavioural functions, this inter-individual variability may also be explained by long-term prior sleep–wake history (Grant & Van Dongen, 2013), albeit only partially (Van Dongen, Baynard, et al., 2004).

It is noteworthy that inter-individual differences in vulnerability to sleep loss were also observed in the sample of the present study, and these systematic differences did not fully average out over participants in each of the study conditions. As such, differences were seen between conditions in the build-up of neurobehavioural performance impairment across the first 5-day period of sleep restriction, even though the experimental conditions were still the same then for all conditions. These idiosyncratic group differences in neurobehavioural outcomes were accounted for in the statistical analyses by means of random effects (Olofsen et al., 2004), and in the corresponding figures (but not in the statistical analyses themselves) by normalization to the first 5-day period of sleep restriction common to all participants.

Notwithstanding the many controls built into this highly standardized, in-laboratory, dose–response study, there are limitations. First, circadian phase was not assessed across the experimental manipulation; it was beyond the scope of the study. However, all groups were exposed to the same sleep restriction schedule beside the sleep dose on the DRI night, light levels were kept below 50 lux during scheduled wakefulness, and neurobehavioural performance and subjective sleepiness assessments were averaged across a fixed time-of-day interval for each study day. Additionally, any study effects on the circadian system should be small and equivalent across groups. Also, there were differences in the magnitude of neurobehavioural impairment between study conditions prior to receiving the randomized sleep dose on the DRI night (Figure 3a,c). Irrespective of these idiosyncratic differences, the magnitude and rate of accumulation of neurobehavioural impairment during the second period of sleep restriction reflected the first period of sleep restriction, in that the build-up of impairment evident during the first sleep restriction period continued during the second period of sleep restriction (Figure 3b,d). Further, in future work, it would be important to assess other cognitive domains, as the dynamic response to sleep loss may be different for other neurocognitive processes (Lo et al., 2019; Smith et al., 2021).

Despite these limitations, our study provided a critical test of the current, sleep–wake homeostasis-based understanding of the impact of sleep loss on neurobehavioural functioning. Our findings revealed that this understanding is incomplete and must be expanded, as the neurobehavioural response to sleep loss is dependent not only on a person's present state of sleep loss, but also fundamentally on the longer-term sleep–wake dynamics that led to the present state. Given the criticality of neurobehavioural functioning for performance and safety, it is important to recognize that the underlying biological processes have long-term dynamics with time constants that exceed the temporal reach of the sleep homeostatic and circadian processes captured by the two-process model (McCauley et al., 2009). Future research should be focused on uncovering the neurobiological and molecular mechanisms underlying these dynamics (Basheer et al., 2007; Eban-Rothschild et al., 2018).

Obtaining adequate sleep has been recognized as an integral component of a healthy lifestyle (Watson et al., 2015). The long-term sleep–wake dynamics uncovered in this study suggest that sleep and health relationships should be considered on a scale of weeks, not days. Generalizing our findings to the public suggests that individuals engaging in sleep–wake–work cycles with chronically insufficient sleep and only intermittent recovery opportunities are operating with a neurobehavioural deficit, which may be markedly underestimated and is increasingly more susceptible to further sleep loss. People may be carrying this neurobehavioural deficit into their professional environment, with attendant consequences for performance and productivity. They may also be impaired, potentially unknowingly, when driving or operating safety-sensitive equipment, thereby posing safety risks to themselves and others. From a health perspective, the pernicious effects of chronic sleep loss may extend beyond neurobehavioural functioning, and increase the risk for metabolic and cardiovascular diseases through dysregulated metabolic and inflammatory processes (Depner et al., 2019; Simpson et al., 2016). However, whether or not the health consequences of chronic sleep loss also exhibit hysteresis remains to be investigated.

Interestingly, the strong influence of sleep–wake history uncovered by our findings also implies that occasional perturbations of a schedule with otherwise regular, adequate sleep may have relatively little impact on neurobehavioural functioning in the long run. If this implication extends to the health consequences of sleep loss as well, then good sleep health should be achievable even by people whose lives are incongruent with the emphasis on day-to-day sleep regularity often found in public messaging campaigns. Considering a more holistic, long-term view on the overall adequacy of sleep across days and weeks, allowing for brief and infrequent deviations from regularity, would help to make public messaging around sleep more inclusive to the many individuals whose sleep schedules are occasionally disrupted by work demands or other circumstances beyond their control.

AUTHOR CONTRIBUTIONS

Siobhan Banks: Investigation; conceptualization; writing – original draft; methodology; writing – review and editing; formal analysis; project administration; data curation. Christopher W. Jones: Writing – review and editing; writing – original draft; formal analysis. Mark E. McCauley: Formal analysis; data curation; writing – review and editing. Jillian Dorrian: Conceptualization; writing – review and editing; formal analysis; data curation. Mathias Basner: Writing – review and editing; formal analysis; data curation. Greg Maislin: Formal analysis; funding acquisition; writing – review and editing. Hans P. A. Van Dongen: Conceptualization; investigation; funding acquisition; writing – original draft; writing – review and editing; methodology; formal analysis; project administration; data curation; supervision; resources. David F. Dinges: Conceptualization; investigation; funding acquisition; writing – original draft; writing – review and editing; methodology; formal analysis; project administration; resources; supervision; data curation.

FUNDING INFORMATION

This research was supported by the National Space Biomedical Research Institute through NASA NC 9–58 (DFD); National Institutes of Health grant NR04281 (DFD); Army Research Office grant W911NF2210223 (HPAVD, MEM); National Institutes of Health T32 grant HL07713 (CWJ); and National Institutes of Health CTRC grant ULRR241340.

CONFLICT OF INTEREST STATEMENT

The authors declare that they have no competing interests.