Interindividual differences in the dynamics of the homeostatic process are trait-like and distinct for sleep versus wakefulness

Transcription

1 J Sleep Res. (2017) 26, Dynamics of sleep-wake homeostasis Interindividual differences in the dynamics of the homeostatic process are trait-like and distinct for sleep versus wakefulness THOMAS RUSTERHOLZ 1,2, LEILA TAROKH 1,2,3, HANS P. A. VAN DONGEN 4,5, * and PETER ACHERMANN 1,6,7,8,9, * 1 Institute of Pharmacology and Toxicology, University of Zurich, Zurich, Switzerland; 2 University Hospital of Child and Adolescent Psychiatry and Psychotherapy, University of Bern, Bern, Switzerland; 3 Department of Psychiatry and Human Behavior, The Alpert Medical School of Brown University, Providence, RI, USA; 4 Sleep and Performance Research Center, Washington State University, Spokane, WA, USA; 5 Elson S. Floyd College of Medicine, Washington State University, Spokane, WA, USA; 6 The KEY Institute for Brain-Mind Research, Department of Psychiatry, Psychotherapy and Psychosomatics, University Hospital of Psychiatry, Zurich, Switzerland; 7 Zurich Center for Interdisciplinary Sleep Research, University of Zurich, Zurich, Switzerland; 8 Zurich Center for Integrative Human Physiology, University of Zurich, Zurich, Switzerland; 9 Neuroscience Center Zurich, University and ETH Zurich, Zurich, Switzerland Keywords ICC, phenotype, mathematical modeling, multiple recordings, sleep/wake regulation, variance components Correspondence Peter Achermann, University of Zurich, Institute of Pharmacology and Toxicology, Section of Chronobiology and Sleep Research, Winterthurerstrasse 190, CH-8057 Zurich, Switzerland. Tel.: ; fax: ; acherman@pharma.uzh.ch *Shared senior authorship. Accepted in revised form 5 November 2016; received 19 October 2016 DOI: /jsr SUMMARY The sleep homeostatic Process S reflects the build-up of sleep pressure during waking and its dissipation during sleep. Process S is modelled as a saturating exponential function during waking and a decreasing exponential function during sleep. Slow wave activity is a physiological marker for non-rapid eye movement (non-rem) sleep intensity and serves as an index of Process S. There is considerable interindividual variability in the sleep homeostatic responses to sleep and sleep deprivation. The aim of this study was to investigate whether interindividual differences in Process S are trait-like. Polysomnographic recordings of 8 nights (12-h sleep opportunities, 22:00 10:00 hours) interspersed with three 36-h periods of sustained wakefulness were performed in 11 healthy young adults. Empirical mean slow wave activity per non-rem sleep episode at episode mid-points were used for parameter estimation. Parameters of Process S were estimated using different combinations of consecutive sleep recordings, resulting in two to three sets of parameters per subject. Intraclass correlation coefficients were calculated to assess whether the parameters were stable across the study protocol and they showed trait-like variability among individuals. We found that the group-average time constants of the build-up and dissipation of Process S were 19.2 and 2.7 h, respectively. Intraclass correlation coefficients ranged from 0.48 to 0.56, which reflects moderate trait variability. The time constants of the build-up and dissipation varied independently among subjects, indicating two distinct traits. We conclude that interindividual differences in the parameters of the dynamics of the sleep homeostatic Process S are trait-like. INTRODUCTION Considerable interindividual variation in sleep physiology has long been observed, especially with regard to the response to sleep deprivation (De Gennaro et al., 2005; Finelli et al., 2001a; Tarokh et al., 2015; Tinguely et al., 2006; Tucker et al., 2007). Despite this, in-depth investigations of interindividual differences in sleep physiology are scarce (Van Dongen et al., 2005). These interindividual differences are considered to be trait-like when they exhibit significant stability over time and are robust to experimental challenges (Van Dongen et al., 2005) or if a strong genetic component can be demonstrated as, for example, in twin studies (Ambrosius et al., 2008; De Gennaro et al., 2008). Studies using repeated sleep deprivation in the same individuals have demonstrated that the waking neurobehavioural response to sleep deprivation (e.g. performance on tasks of cognitive function) is stable, robust and unique to an 171

2 172 T. Rusterholz et al. individual. Thus, these studies revealed trait-like differential vulnerability to sleep loss (Rupp et al., 2012; Van Dongen et al., 2004a). The degree to which various sleep parameters are trait-like has also been examined using a repeated sleep deprivation experiment (Tucker et al., 2007). In this study, intraclass correlation coefficients (ICCs) were used as a measure of intra-individual stability and interindividual variability, with an ICC of 1 indicating a perfect trait. It was found that ICCs for sleep parameters ranged from 0.36 to 0.89, with the highest ICCs found in the duration of slow wave sleep (SWS) and in delta power of non-rapid eye movement (NREM) sleep (Tucker et al., 2007). Sleep variables did not vary among subjects in an independent manner, as approximately 60% of their combined variance clustered in three trait dimensions essentially representing sleep duration, sleep intensity and sleep discontinuity (Tucker et al., 2007). In other studies, spectra of the NREM and rapid eye movement (REM) sleep electroencephalograms (EEG) have been shown to display trait-like features across multiple polysomnographic recordings at baseline and following sleep deprivation (Adamczyk et al., 2015; Ambrosius et al., 2008; Buckelm uller et al., 2006; De Gennaro et al., 2008; Finelli et al., 2001a; Tan et al., 2001; Tarokh et al., 2015). Sleep architecture (i.e. the distribution of different sleep stages across the sleep period) (Buckelm uller et al., 2006) and the minute-to-minute time course of slow wave activity during sleep (Achermann and Borbely, 1990) appear to be less traitlike. From these findings, it remains unclear whether the underlying sleep homeostatic process, Process S, is trait-like. Sleep homeostasis describes the sleep wake dependent aspect of sleep regulation. Homeostatic sleep pressure increases monotonically during waking and decreases monotonically during sleep. Slow wave activity (SWA; EEG spectral power in the Hz range) is a physiological marker for NREM sleep intensity and serves as an index of sleep homeostasis. The decline of SWA across consecutive NREM sleep episodes is predominantly exponential (Achermann and Borbely, 2011; Borbely, 1982; Daan et al., 1984). Additionally, mean SWA increases as a function of prior wake duration, especially in the first NREM sleep episode (Beersma et al., 1987; Daan et al., 1984). These dynamics are captured in the homeostatic Process S of the two-process model of sleep regulation (Borbely, 1982; Daan et al., 1984), which represents the accumulation of sleep pressure (propensity) during waking and its dissipation during sleep. In the two-process model, Process S is modelled with a saturating exponential function during waking and an exponential decline during sleep. In the formulation of the two-process model, it has been assumed that the time constant of the decay of Process S is the same during a baseline night of sleep and during recovery sleep after prolonged wakefulness. This has been confirmed empirically (Dijk et al., 1990, 1991; Rusterholz et al., 2010). However, this result was based mainly on modelling of pooled data, and therefore does not allow for inferences about the stability of the dynamics of Process S within subjects. We have established previously a reliable method for estimating the parameters of Process S (time constants and asymptotes) in individual recordings, and demonstrated considerable interindividual variation in all parameters (Rusterholz et al., 2010). In this work, the parameters of Process S were assumed to be stable within (adult) individuals over time. To date, however, this assumption has not been systematically tested experimentally. The aim of the present study was to investigate, using data from a laboratory study, whether or not individual differences in the parameters of Process S are trait-like. We analysed polysomnographic (PSG) recordings of healthy young adults who underwent a laboratory protocol with repeated exposure to sleep deprivation (Tucker et al., 2007). Using statistical methodology developed to dissociate variance components in data sets with repeated measures in different individuals (Olofsen et al., 2004; Van Dongen et al., 2004b), trait aspects of Process S dynamics were evaluated by comparing the interand intra-individual variability of the parameter estimates. METHODS Subjects Eleven subjects [six men, five women; mean age standard deviation (SD): years; six African American, five Caucasian] were included in the analysis. All subjects were screened carefully to be healthy and free of medications (other than contraceptives), and they reported to be good sleepers with habitual sleep durations between 7 and 9 h per night. Subjects were required to adhere to their habitual bedtimes during the 7 days prior to the experiment and had to abstain from caffeine, tobacco, alcohol and drugs for these 7 days and during the experiment. Study design Subjects underwent a protocol with repeated sleep deprivation in a highly controlled laboratory environment. PSG recordings from eight sleep periods (S1 S8; 12 h time in bed, 22:00 10:00 hours), interspersed with three 36-h periods of total sleep deprivation, were analysed for each subject. Fig. 1 depicts the laboratory protocol. Further details of the study can be found in Tucker et al. (2007); we used a subset of data from this study comprising only subjects who had complete data for the entire protocol or at least the first six sleep episodes (S1 S6; see Fig. 1). Polysomnography Four EEG derivations (Fz, C3, C4 and Oz) were recorded against linked mastoid (LM). Sleep stages were scored for 30-s epochs according to the criteria of Rechtschaffen and Kales (1968). Power density spectra were determined for 30-s epochs (fast Fourier transform, average of 10 4-s epochs overlapping by 1 s, cosine-tapered). Artefacts were excluded semiautomatically when power in the SWA and

3 Stability of the dynamics of sleep homeostasis 173 Figure 1. Laboratory protocol. S1 S8: polysomnographic (PSG) recordings of 12-h nocturnal sleep opportunities; time in bed from 22:00 to 10:00 hours. W: scheduled 12-h waking periods from 10:00 to 22:00 hours. DEP: scheduled periods of 36-h total sleep deprivation from 10:00 to 22:00 hours the next day. Dashed lines: different sets of consecutive sleep recordings (data blocks) used for parameter estimation (e.g. using recordings S2 S4). S1 was considered an adaptation night and was not used for the present analyses Hz frequency bands exceeded predefined thresholds based on a moving average determined over s epochs (Finelli et al., 2001b). NREM and REM sleep episodes were determined according to the criteria of Feinberg and Floyd (1979), with the following modification: subjects did not necessarily sleep through the night and, in some cases, we observed several wake periods during the scheduled 12-h sleep opportunity. Sustained waking periods longer than or equal to 10 min were treated as interspersed waking episodes with accompanying increases of the homeostatic Process S (see Fig. S1). Parameter estimation The homeostatic Process S was modelled by a saturating exponential function during waking and a decreasing exponential function during sleep (Borbely, 1982; Daan et al., 1984; Rusterholz et al., 2010): SðtÞ ¼ðS SO LAÞexpð t s d ÞþLA during sleep (1) SðtÞ ¼ðS WO UAÞexpð t s i ÞþUA during waking (2) Here S SO and S WO are the levels of S at sleep onset and wake onset, respectively. LA is the lower asymptote and UA the upper asymptote; s d is the time constant of the exponential decline during sleep, and s i is the time constant of the saturating exponential increase during waking. The focus of the analysis on the asymptotes was on the distance between them. Mean SWA (derivation C3-LM) of NREM sleep episodes (stages 2 4) at episode mid-points served as input for leastsquares parameter estimation. In three subjects, derivation C4-LM had to be used, because of incomplete C3-LM data. SWA values were normalized within each subject using the mean SWA in the first 6 h of the baseline sleep recording (S2; adaptation night S1 was not used). Within these first 6 h of baseline sleep, no subject experienced a waking episode longer than or equal to 10 min. In order to assess whether the build-up and dissipation of sleep pressure are stable and unique to an individual, we estimated the parameters of Process S for individuals using Figure 2. Illustration of simulations of Process S. Black dots denote mean slow wave activity (SWA) per non-rapid eye movement (NREM) sleep episode at episode midpoints for subject F. The curves represent simulations of Process S with optimized parameters for the three data blocks S2 S4, S4 S6 and S6 S8.

4 174 T. Rusterholz et al. the method of Rusterholz et al. (2010). Parameters were estimated initially with average data. The resulting time constants served to constrain the individual time constants to a physiological meaningful range. For further details, see Supporting Information. Due to recording issues (poor data quality or missing recordings for S7 or S8), not all nights of the 11 subjects could be used. Data of the entire protocol (S2 S8) were available for six subjects. These subjects data were used to estimate group-average parameters of Process S. Recordings of S2 S6 were available for all 11 subjects. Recordings of S2 S6, or S2 S8 where available, were used to estimate subject-specific parameters of Process S for all 11 subjects. To evaluate the trait-like aspects of Process S, we split the sleep recordings into three data blocks, each containing 3 consecutive nights of PSG recordings (S2 S4, S4 S6 and S6 S8; see Fig. 1). Each data block started with a night of sleep after a scheduled 12-h waking period, followed by a recovery night after a scheduled 36-h sleep deprivation period, and ended with a night of sleep after another scheduled 12-h waking period. Basing our analyses on three sleep recordings, including the night after recovery from Table 1 Grand means, variance components and ICCs for s i, s d and UA-LA in the primary variance components analysis (1) and different secondary analyses (2a 2d; see Statistical analyses) to verify robustness of the findings. The primary results (using all data blocks) are shown in bold type Parameter Analysis Grand mean Var(bs) Var(ws) ICC s i (h) ( ) 2a ( ) 2b ( ) 2c ( ) 2d ( ) s d (h) ( ) 2a ( ) 2b ( ) 2c ( ) 2d ( ) UA-LA (%) ( ) 2a ( ) 2b ( ) 2c ( ) 2d ( ) Values shown are estimate standard error, and parentheses indicate 95% confidence intervals for the intraclass correlation coefficient (ICC). Var(bs), between-subject variance; Var(ws), within-subject variance. Figure 3. Grand mean estimates (estimate standard error) of time constants (s i, s d ) and the difference between the asymptotes (UA-LA) for the primary variance components analysis (analysis 1) and for the secondary analyses (2a 2d) conducted to verify robustness (see Statistical analyses). For comparison, values estimated with the same approach in a different data set (Rusterholz et al., 2010) (lit) are also shown.

5 Stability of the dynamics of sleep homeostasis 175 sleep deprivation, yielded stable parameter estimates. Fits of Process S for an example subject are depicted in Fig. 2. Statistical analyses The individual subjects model parameters estimated for each of the data blocks time constants s i and s d and the distance between the asymptotes UA-LA were analysed across subjects using variance components analysis to dissociate between- from within-subject variance. The variance components analysis was implemented in the following way: Analysis 1: as the primary analysis, the variance components analysis was implemented using random-effect regression with each model parameter as the dependent variable regressed separately against a random effect over subjects on the intercept. Analysis 2a: to verify robustness of the findings, as a secondary analysis, the primary analysis was repeated controlling for data blocks to account for any order effects in the data. Analysis 2b: another secondary analysis repeated analysis 1, but excluded subject I (see Fig. S2). Analysis 2c: yet another secondary analysis repeated analysis 1, but used only the six subjects who had complete data. Analysis 2d: a secondary analysis repeated analysis 2c discarding the S4 S6 block, so that there was no overlap between the blocks and therefore no potential for interdependence of the parameter estimates across blocks. For each of these analyses the grand mean over subjects and blocks was extracted to compare to previously published group-average data. To quantify the degree to which each parameter was stable across data blocks, the ICC was calculated as the between-subject variance divided by the between- plus within-subject variances (Van Dongen et al., 2004b). RESULTS Table 1 and Fig. 3 show the grand mean estimates of s i, s d and UA-LA for the primary variance components analysis (analysis 1) and for the secondary analyses (2a 2d) conducted to verify robustness (see above). Across these different analyses, the grand means were consistent and within less than 1 standard error of each other, indicating that the findings are robust. The grand mean values are close to those reported earlier (Rusterholz et al., 2010) based on estimations performed on a different data set (Finelli et al., 2001b). Figure 4 shows the estimates of the parameters for each data block of each individual subject. As seen in Fig. 4, for each of the time constants s i and s d, and for the difference between the asymptotes UA-LA, there were substantial interindividual differences, whereas the values were relatively stable for each individual over data blocks. Table 1 and Fig. 5 shows the corresponding between- and within-subject Figure 4. Estimates of the parameters s i, s d, and UA-LA for each data block of each individual subject (A K). In each panel, subjects were rank-ordered by the overall value that best represents their individual trait. variances and the ICCs. The ICCs, ranging from for s d to for s i, showed moderate stability of the interindividual differences. This indicates that these parameters of the

6 176 T. Rusterholz et al. Figure 5. Intraclass correlation coefficients (ICCs) of the parameters s i, s d, and UA-LA of the primary variance components analysis (1) and different secondary analyses (2a 2d; see Statistical analyses). dynamics of sleep homeostasis were stable and may reflect subject-specific traits. In each panel of Fig. 4 the subjects are rank-ordered by overall value that best represents their individual trait (estimated best linear unbiased predictor in analysis 1; see Supporting Information). The order of the subjects in the three panels of Fig. 4 varies from one parameter to another. The rank-order correlation between s i and s d across subjects was q = (P = 0.979), suggesting that the stable interindividual differences in these two parameters are not interdependent and may reflect two separate traits. The rank-order correlation between s i and UA-UL was q = (P = 0.013) and between s d and UA-UL was q = (P = 0.223). The rank-order correlation between s i and UA UL, which was the only rank-order correlation that reached statistical significance, is driven in part by poor separability of these parameters in the parameter estimation process. Although the result suggests that these parameters may reflect a single underlying trait, this interpretation should be considered tentative. DISCUSSION Sleep homeostasis is reflected in the parameters of the homeostatic process, Process S [see Rusterholz et al. (2010) for a discussion], and can be modelled using SWA in the sleep EEG. We conducted the first investigation examining whether the parameters of Process S are stable and unique within individuals. Our ICC analyses (Table 1; Fig. 5) revealed within-subject stability across baseline and sleep deprivation periods and considerable betweensubject variability, yielding ICCs in the moderate range according to published criteria (Landis and Koch, 1977). Thus, we conclude that the parameters of Process S are trait-like. In line with our finding that the dissipation of homeostatic sleep pressure differs among individuals in a trait-like manner, Schmidt et al. (2009) reported a faster dissipation of sleep pressure in morning than evening types. Furthermore, there is evidence that sleep regulation is under genetic control. In mice, systematic differences in sleep homeostasis have been shown between different strains (Franken et al., 2001). In humans, a faster homeostatic dissipation seems to occur in individuals with the period circadian clock 3 (PER3) 5/5 polymorphism compared to individuals with PER3 4/4 (Viola et al., 2007), although time constants were not reported for this comparison. Other genotypes implicated in human sleep/wake regulation, such as polymorphisms of adenosine deaminase (ADA) (Bachmann et al., 2012), may also affect the dissipation time constant and need further investigation. In our study, we could not determine the relative contribution of genes to the betweensubjects variance in parameters of Process S; twin studies will be needed to address this question. To our knowledge, there has been only one study of individual differences in the build-up of homeostatic pressure during wakefulness based on SWA in the preceding and subsequent sleep periods (Rusterholz et al., 2010), and this study did not investigate the stability of the model parameters. However, studies of waking correlates of the homeostatic build-up during extended wakefulness (after controlling for circadian rhythm effects) have documented trait-like variability of sleep latency in the multiple sleep latency test (MSLT) (Roth et al., 1997) and cognitive deficits in neurobehavioural performance tasks (Rupp et al., 2012; Van Dongen et al., 2004a). It has been argued that, from such findings, stable interindividual differences in the build-up of

7 Stability of the dynamics of sleep homeostasis 177 homeostatic pressure may be inferred (Van Dongen et al., 2012). Furthermore, there is evidence from neurobehavioural performance data that the build-up of homeostatic pressure during extended wakefulness may be under genetic control (Satterfield et al., 2015). Our analysis revealed that the two time constants of Process S may reflect two distinct traits. To provide a context for this finding, consider that a long time constant of the homeostatic decline indicates a slow dissipation of sleep pressure, while a long time constant of the homeostatic buildup may point to a better capacity of an individual to cope with sleep deprivation or sleep restriction. This conjecture is supported by changes in the build-up of Process S across development, with homeostatic sleep pressure accumulating more slowly during wakefulness with increasing age, with no accompanying change to the dissipation of Process S (Forsman and Van Dongen, 2013; Jenni et al., 2005). Based on our finding that the parameters of Process S are trait-like, we would also expect to find that sleep duration and the ability to withstand sleep pressure are trait-like. Indeed, sleep duration has been found to be moderately trait-like (Tucker et al., 2007) and under genetic control (De Castro, 2002; Hublin et al., 2013; Partinen et al., 1983), and individuals who consistently have either short or long sleep have been reported in the literature (Aeschbach et al., 2001). Individual differences in neurobehavioural performance during sleep deprivation appear to be independent of individual differences in sleep duration (and other sleep variables such as the amount of NREM sleep; e.g. Møst et al. (2003)). Our present results contribute to the evidence that sleep duration and vulnerability to sleep loss are two distinct traits (Grant and Van Dongen, 2013). Seemingly in contradiction to the above findings are the results of studies in habitual short and long sleepers. Comparing the dynamics of Process S in short and long sleepers, Aeschbach et al. (2001) did not find differences between these groups neither for the build-up (as assessed based on theta activity in the wake EEG) nor for the decline (as assessed with SWA during recovery sleep). One caveat is that the statistical comparisons were based on asymptotic 95% confidence intervals around the means and not on individual parameter estimates. Our finding that the dissipation and build-up time constants vary among individuals independently casts new light on this issue. That is, another possibility for manifesting short sleep may involve fast dissipation and/or slow build-up time constants. Furthermore, the relatively small sample size of the study of Aeschbach et al. (2001) (nine short sleepers, eight long sleepers) may have hampered detection of differences between the groups. In conclusion, our method to estimate the parameters of Process S in individual recordings (Rusterholz et al., 2010) allowed us to investigate the stability of the homeostatic process in individual subjects. We demonstrated for the first time that the parameters of the homeostatic process (time constants and asymptotes) are trait-like. Furthermore, we found evidence that the time constants for dissipation and build-up constitute two independent traits. ACKNOWLEDGEMENTS This was not an industry-supported study. The study was supported by the Swiss National Science Foundation grants and 32003B_ (to PA) and NIH grant R01HL (to HVD). We thank Dr Roland D urr for computational support. AUTHOR CONTRIBUTIONS HVD designed the study and performed the data acquisition, TR, HVD and PA performed the data analysis and TR, LT, HVD and PA wrote the manuscript. CONFLICT OF INTEREST The authors declare no conflicts of interest. REFERENCES Achermann, P. and Borbely, A. A. Simulation of human sleep: ultradian dynamics of electroencephalographic slow-wave activity. J. Biol. Rhythms, 1990, 5: Achermann, P. and Borbely, A. A. Sleep homeostasis and models of sleep regulation. In: M. Kryger, T. Roth and W. Dement (Eds) Principles and Practices of Sleep Medicine. Elsevier Saunders, Missouri, 2011: Adamczyk, M., Ambrosius, U., Lietzenmaier, S., Wichniak, A., Holsboer, F. and Friess, E. Genetics of rapid eye movement sleep in humans. Transl. Psychiatry, 2015, 5: e598. Aeschbach, D., Postolache, T. T., Sher, L., Matthews, J. R., Jackson, M. A. and Wehr, T. A. Evidence from the waking electroencephalogram that short sleepers live under higher homeostatic sleep pressure than long sleepers. Neuroscience, 2001, 102: Ambrosius, U., Lietzenmaier, S., Wehrle, R. et al. Heritability of sleep electroencephalogram. Biol. Psychiatry, 2008, 64: Bachmann, V., Klaus, F., Bodenmann, S. et al. Functional ADA polymorphism increases sleep depth and reduces vigilant attention in humans. Cereb. Cortex, 2012, 22: Beersma, D. G. M., Daan, S. and Dijk, D. J. Sleep intensity and timing: a model for their circadian control. In: G. A. Carpenter (Ed) Some Mathematical Questions in Biology Circadian Rhythms. The American Mathematical Society, Providence, RI, 1987: Borbely, A. A. A two process model of sleep regulation. Hum. Neurobiol., 1982, 1: Buckelm uller, J., Landolt, H.-P., Stassen, H. H. and Achermann, P. Trait-like individual differences in the human sleep electroencephalogram. Neuroscience, 2006, 138: Daan, S., Beersma, D. G. M. and Borbely, A. A. Timing of human sleep: recovery process gated by a circadian pacemaker. Am. J. Physiol., 1984, 246: R161 R178. De Castro, J. M. The influence of heredity on self-reported sleep patterns in free-living humans. Physiol. Behav., 2002, 76: De Gennaro, L., Ferrara, M., Vecchio, F., Curcio, G. and Bertini, M. An electroencephalographic fingerprint of human sleep. Neuroimage, 2005, 26: De Gennaro, L., Marzano, C., Fratello, F. et al. The electroencephalographic fingerprint of sleep is genetically determined: a twin study. Ann. Neurol., 2008, 64:

8 178 T. Rusterholz et al. Dijk, D. J., Brunner, D. P. and Borbely, A. A. Time course of EEG power density during long sleep in humans. Am. J. Physiol., 1990, 258: R650 R661. Dijk, D. J., Brunner, D. P. and Borbely, A. A. EEG power density during recovery sleep in the morning. Electroencephalogr. Clin. Neurophysiol., 1991, 78: Feinberg, I. and Floyd, T. C. Systematic trends across the night in human sleep cycles. Psychophysiology, 1979, 16: Finelli, L. A., Achermann, P. and Borbely, A. A. Individual fingerprints in human sleep EEG topography. Neuropsychopharmacology, 2001a, 25: S57 S62. Finelli, L. A., Borbely, A. A. and Achermann, P. Functional topography of the human nonrem sleep electroencephalogram. Eur. J. Neurosci., 2001b, 13: Forsman, P. M. and Van Dongen, H. P. A. Age and individual determinants of cognitive effects of sleep loss. In: C. Kushida (Ed) Encyclopedia of Sleep. Academic Press, Waltham, MA, 2013: Franken, P., Chollet, D. and Tafti, M. The homeostatic regulation of sleep need is under genetic control. J. Neurosci., 2001, 21: Grant, D. A. and Van Dongen, H. P. A. Individual differences in sleep duration and responses to sleep loss. In: P. Shaw, M. Tafti and M. Thorpy (Eds) Genetic Basis of Sleep and Sleep Disorders. Cambridge University Press, Cambrige, 2013: Hublin, C., Partinen, M., Koskenvuo, M. and Kaprio, J. Genetic factors in evolution of sleep length a longitudinal twin study in Finnish adults. J. Sleep Res., 2013, 22: Jenni, O. G., Achermann, P. and Carskadon, M. A. Homeostatic sleep regulation in adolescents. Sleep, 2005, 28: Landis, J. R. and Koch, G. G. The measurement of observer agreement for categorical data. Biometrics, 1977, 33: Møst, E. I. S., Baynard, M. D., Fox, C. G. et al. On the relationship between interindividual differences in performance impairment from sleep loss and interindividual differences in sleep architecture. Sleep-Wake Res. Netherlands, 2003, 14: Olofsen, E., Dinges, D. F. and Van Dongen, H. P. A. Nonlinear mixed-effects modeling: individualization and prediction. Aviat. Space Environ. Med., 2004, 75: A134 A140. Partinen, M., Kaprio, J., Koskenvuo, M., Putkonen, P. and Langinvainio, H. Genetic and Environmental determination of human sleep. Sleep, 1983, 6: Rechtschaffen, A. and Kales, A. A Manual of Standardized Terminology, Techniques and Scoring System for Sleep Stages of Human Subjects. National Institutes of Health, Bethesda, MD, Roth, T., Roehrs, T. and Rosenthal, L. Daytime sleepiness in normals: individual differences in sleep need not ability to fall asleep. In: O. Hayaishi and S. Inoue (Eds) Sleep and Sleep Disorders: From Molecule to Behavior. Academic Press, Tokyo, 1997: Rupp, T. L., Wesensten, N. J. and Balkin, T. J. Trait-like vulnerability to total and partial sleep loss. Sleep, 2012, 35: Rusterholz, T., D urr, R. and Achermann, P. Interindividual differences in the dynamics of sleep homeostasis. Sleep, 2010, 33: Satterfield, B. C., Wisor, J. P., Field, S. A., Schmidt, M. A. and Van Dongen, H. P. A. TNFa G308A polymorphism is associated with resilience to sleep deprivation-induced psychomotor vigilance performance impairment in healthy young adults. Brain Behav. Immun., 2015, 47: Schmidt, C., Collette, F., Leclercq, Y. et al. Homeostatic sleep pressure and responses to sustained attention in the suprachiasmatic area. Science, 2009, 324: Tan, X., Campbell, I. G. and Feinberg, I. Internight reliability and benchmark values for computer analyses of non-rapid eye movement (NREM) and REM EEG in normal young adult and elderly subjects. Clin. Neurophysiol., 2001, 112: Tarokh, L., Rusterholz, T., Achermann, P. and Van Dongen, H. P. A. The spectrum of the non-rapid eye movement sleep electroencephalogram following total sleep deprivation is trait-like. J. Sleep Res., 2015, 24: Tinguely, G., Finelli, L. A., Landolt, H. P., Borbely, A. A. and Achermann, P. Functional EEG topography in sleep and waking: state-dependent and state-independent features. Neuroimage, 2006, 32: Tucker, A. M., Dinges, D. F. and Van Dongen, H. P. A. Trait interindividual differences in the sleep physiology of healthy young adults. J. Sleep Res., 2007, 16: Van Dongen, H. P. A., Baynard, M. D., Maislin, G. and Dinges, D. F. Systematic interindividual differences in neurobehavioral impairment from sleep loss: evidence of trait-like differential vulnerability. Sleep, 2004a, 27: Van Dongen, H. P. A., Olofsen, E., Dinges, D. F. and Maislin, G. Mixed-model regression analysis and dealing with interindividual differences. Methods Enzymol., 2004b, 384: Van Dongen, H. P. A., Vitellaro, K. M. and Dinges, D. F. Individual differences in adult human sleep and wakefulness: leitmotif for a research agenda. Sleep, 2005, 28: Van Dongen, H. P. A., Bender, A. M. and Dinges, D. F. Systematic individual differences in sleep homeostatic and circadian rhythm contributions to neurobehavioral impairment during sleep deprivation. Accid. Anal. Prev., 2012, 45 (Suppl.): Viola, A. U., Archer, S. N., James, L. M. et al. PER3 polymorphism predicts sleep structure and waking performance. Curr. Biol., 2007, 17: SUPPORTING INFORMATION Additional Supporting Information may be found online in the supporting information tab for this article: Figure S1. Representative recording of a night with 12-h sleep opportunity. Top: slow wave activity (SWA) of artifactfree 30-s epochs of sleep (black line), mean SWA per NREM sleep episode at episode midpoint (grey points), and fitted Process S (grey line). Middle: NREM sleep (grey), REM sleep (white), and waking (black) episodes. Bottom: hypnogram (W: waking; R: REM sleep; 1 4: NREM sleep stages 1 to 4). Figure S2. Simulations of Process S in subject I. The empirical data (SWA) of the first data block (S2 S4) were at a lower level compared to those of the remaining nights. This affected parameter estimation in the second data block (S4 S6) in particular, as both levels contributed to the estimation in that data block. Subject I showed by far the largest discrepancy between empirical data and simulations, and may be an outlier. For figure details, see Figure 2 caption. Figure S3. Histogram of REM sleep latencies (up to 30 min) after initial sleep onset and after consolidated interspersed waking periods ( 10 min). Table S1. Average parameters and standard deviations obtained from parameter estimation in individuals, treating SOREM sleep episodes either as sleep or as waking, including only data blocks with at least one SOREM sleep episode. The two assumptions were compared with paired t- tests.