Effects of sleep restriction on the sleep electroencephalogram of adolescents

Abstract

Study Objectives

This report describes findings from an ongoing longitudinal study of the effects of varied sleep durations on wake and sleep electroencephalogram (EEG) and daytime function in adolescents. Here, we focus on the effects of age and time in bed (TIB) on total sleep time (TST) and nonrapid eye movement (NREM) and rapid eye movement (REM) EEG.

Methods

We studied 77 participants (41 male) ranging in age from 9.9 to 16.2 years over the 3 years of this study. Each year, participants adhered to each of three different sleep schedules: four consecutive nights of 7, 8.5, or 10 h TIB.

Results

Altering TIB successfully modified TST, which averaged 406, 472 and 530 min on the fourth night of 7, 8.5, and 10 h TIB, respectively. As predicted by homeostatic models, shorter sleep durations produced higher delta power in both NREM and REM although these effects were small. Restricted sleep more substantially reduced alpha power in both NREM and REM sleep. In NREM but not REM sleep, sleep restriction strongly reduced both the all-night accumulation of sigma EEG activity (11–15 Hz energy) and the rate of sigma production (11–15 Hz power).

Conclusions

The EEG changes in response to TIB reduction are evidence of insufficient sleep recovery. The decrease in sigma activity presumably reflects depressed sleep spindle activity and suggests a manner by which sleep restriction reduces waking cognitive function in adolescents. Our results thus far demonstrate that relatively modest TIB manipulations provide a useful tool for investigating adolescent sleep biology.

Statement of Significance.

This dose-response study systematically altered time in bed duration to examine the effects of sleep reduction on adolescents’ sleep electroencephalogram (EEG). Sleep restriction produced a small increase in delta (1–4 Hz) EEG power and much larger decreases in alpha (8–11 Hz) power (in both nonrapid eye movement (NREM) and rapid eye movement (REM)) and in NREM sigma (11–15 Hz) power. These findings demonstrate strong effects of modest sleep restriction on adolescent sleep EEG and may help unravel the biological significance of the marked EEG changes in adolescent sleep. The substantial reduction of sigma power, a reflection of sleep spindle activity, may point to a means by which sleep loss affects adolescents’ cognitive function.

Introduction

Our laboratory has proposed that the human brain undergoes a massive adolescent reorganization driven by synaptic pruning [1] and that the sleep electroencephalogram (EEG) provides a valuable, noninvasive index of these maturational changes [2]. We have been investigating these changes longitudinally since 2002. Our initial studies established the maturational trajectory of nonrapid eye movement (NREM) sleep slow-wave EEG power which falls massively during adolescence [2, 3]. Other laboratories have also found this adolescent decline in delta EEG power [4–6]. We demonstrated that the steepest rate of slow-wave power decline occurs between ages 12 and 16.5 years and that this follows by about 1 year the most rapid increase in pubertal maturation [7]. These studies also documented with EEG recording the 10 min/year decline in self-selected weeknight sleep duration across adolescence [8] previously found in a large population survey [9]. In addition, our longitudinal EEG recordings established that this adolescent decline in weeknight sleep durations is entirely produced by decreases in NREM sleep.

In addition to basic science implications, the changing sleep patterns across adolescence touch on an important public health concern: how much sleep do adolescents need at different ages? To address this issue, we employed a dose–response design to investigate the relations between prior sleep duration and waking performance across the teenage years. This dose–response paradigm also allowed us to examine the effects of varied sleep durations on the nighttime sleep EEG, which is the main subject of the present report.

Homeostatic/recuperative models of slow-wave sleep [10, 11] predict that restricting sleep should increase sleep need and increase slow-wave EEG power during NREM sleep. However, in adults even sleep restriction by as much as 4 h per night does not increase slow-wave power on subsequent nights [12–14]. We recently hypothesized that children may be more susceptible to the effect of sleep loss than are adults [15] and might, therefore, respond with increased delta power to small increases in prior waking. However, the first year of this study did not show a delta power increase in response to sleep restriction [15].

With additional longitudinal data, we now revisit the effects of sleep duration on the sleep EEG in adolescents. Rather than limiting our analyses to delta frequencies in NREM sleep, we report the effects of varied sleep duration on EEG power in frequencies from 0.3 to 23 Hz in both NREM and REM sleep.

Methods

Subjects

Seventy-seven participants (41 male, 36 female) enrolled in this 3-year longitudinal study. Over the 3 years of the study, participant ages ranged from 9.9 to 16.2 years of age (mean ± SD = 13.2 ± 1.4).

Recruitment and enrollment details have been previously published [16]. Briefly, subjects were selected from Davis, Dixon, and Woodland, California, which are all located within 20 miles of the UC Davis Sleep Lab. Interviews with participants’ parent(s) screened for the following exclusion criteria: diagnosed sleep disorder, diagnosed mental illness or neurological disorder, epilepsy, head injury with loss of consciousness, and symptoms persisting longer than 24 h, a Sleep Disturbance Scale for Children t-score >70, use of any medication that would affect sleep, and visual problems or manual dexterity problems that would interfere with daytime performance testing. Parents provided written informed consent, and study participants provided assent. Participants received monetary compensation for completing each phase of the study. The UC Davis Institutional Review Board reviewed and approved all procedures used in this study.

Experimental design

Each year, participants completed three different week-long assigned sleep schedules. Each schedule started with three nights with 8.5 h in bed, followed by four consecutive nights of either 7, 8.5, or 10 h in bed. Time in bed (TIB) was altered by changing bedtime; rise time was each subject’s habitual weekday rise time. Participants’ school and social activity prevented randomizing the order of the different TIB schedules. Instead, we were fairly successful in balancing the order that participants completed the TIB schedules. For example, across the 3 years of the study, the percentage of which TIB schedule was completed first was 36.5%, 31.5%, and 32% for 7, 8.5, and 10 h, respectively. Participants wore actigraphy watches to document their adherence to the assigned TIB schedule. If actigraphy or EEG recording (see below) indicated that the participant deviated from the assigned schedule by more than 1 h, the week would be rescheduled. Immediately after completing each four-night sleep schedule, participants spent a weekend day in the lab for performance tests. Daytime testing results and their relations to sleep EEG changes will be presented in a future report.

EEG recordings

On the second and fourth nights of the four consecutive nights on each assigned TIB schedule, we recorded all-night sleep EEG in the participants’ homes in their habitual sleep environment. Trained technicians applied the following electrodes: F3, F4, C3, C4, P3, P4, O1, O2, A1, A2, LOC, ROC, forehead, two chin locations, and reference and ground electrodes on the scalp and face. After starting all-night EEG recordings on Grass Aura ambulatory recorders (400 Hz digitization rate), technicians left the home and did not monitor the recording during the night. Additional recording details have been previously published [16]. Each night’s recorded data was stored on a flashcard and was downloaded on a lab computer after the night.

EEG analysis

All-night EEG recordings were scored for sleep stages using PASS PLUS (Delta Software, St. Louis) to display the digitized data. Before scoring EEG stages, scorers briefly determined whether channel C3-A2 or C4-A1 had fewer artifacts. Then the scorer scored each 20-s epoch as either wake, NREM 1, NREM 2, NREM 3, or REM based on 2007 AASM visual scoring standards [17]. The epochs that had artifacts were also identified. Sleep cycles were calculated based on Feinberg and Floyd criteria [18]. Four trained scorers and one senior scientist completed and checked all scoring. Nights with less than 5 h EEG recording were excluded from further analysis.

Fast Fourier Transform (FFT) was performed on artifact-free epochs using 5.12 s Welch tapered windows with 2.62-s overlap. Two different sets of FFT frequency bands were used: an initial analysis using standard frequency bands (delta 1.08–4.00 Hz, theta 4.00–7.91 Hz, alpha 7.91–11.04 Hz, sigma 11.04–14.94 Hz, and low beta 14.94–22.95 Hz) and more detailed analysis with 0.195 Hz bands between 4 and 16 Hz. We present FFT energy as a measure of how much activity in a frequency band accumulated across all NREM or REM sleep of the entire night, and we present FFT power as a measure of average activity in a frequency band. Because sleep duration differed for the three TIB conditions, we report power in the first 5 h of NREM, a duration common to all three TIB conditions. For REM sleep, we report power averaged over REM periods 2 and 3 because participants keeping the 7-h TIB schedule often did not have more than three complete REMPs and, for this age group, the first REMP is frequently very short or even skipped completely [19, 20]. In the current dataset, 27% of the nights had a skipped first REM period with an abnormally long cycle1. In these cases, cycle 1 was broken into two cycles during the stage N2 sleep that separated two clear delta power peaks. For comparison to REM periods 2 and 3 data, power averaged over NREM periods 2 and 3 was also analyzed.

The EEG data described in this manuscript will be shared in response to reasonable requests to the corresponding author.

Statistical analysis

As in the previous study [15], SAS mixed-effects analysis was used to evaluate the effects of the three TIB sleep schedules and age on sleep duration measures, EEG power and EEG energy. Mixed-effects analysis is particularly suited for longitudinal data because it accounts for the correlation between multiple observations on the same subject [21]. We treated TIB as a categorical measure and age as a continuous measure. The linear mixed-effects models included a random intercept and did not include any random slopes; the effects of TIB and age were assumed to be the same across participants. Despite our previous findings on complex age-related changes in sleep EEG [2, 3, 22], the analyses in this report modeled a linear change in age. As stated above, we analyzed data in fine and broad frequency bands. The fine band analysis involved 73 analyses each for 5 h NREM, REMP 2 & 3, and NREMP 2 & 3 to determine TIB effects on the power spectra. For these analyses, we report the range of frequencies with significant (α = 0.01) TIB effects. For other analyses, we report exact p values unless p is <0.0001. Although the graphs included in this report show TIB effects on power or energy, we used log power and log energy as the dependent variables in statistical analyses because the log data were more normally distributed. For the graphs, TIB data were averaged over multiple years of recordings, but TIB, age, and their interaction were included as factors in all mixed-effects analyses.

Two different analyses examined TIB effects on the distribution of delta power over the night. One method used linear mixed-effects analysis to evaluate the percentage of delta energy occurring in the first NREM period. The other method used non-linear mixed effect analysis [23] to evaluate the parameters of the equation that describes the exponential decline of delta across the night.

Results

The results presented below are based on the fourth night of each four-night TIB schedule unless specified otherwise.

Effect of TIB and age on total sleep time (TST) and sleep stages

As shown in Table 1, reducing TIB significantly (F_2,149 = 1559, p < 0.0001) decreased TST duration and did so in a nearly linear manner with TST durations at 7 and 10 h TIB both differing significantly (p < 0.0001 for both) from TST at 8.5 h TIB. Reducing TIB significantly decreased both NREM (F_2,149 = 795, p < 0.0001) and REM durations (F_2,149 = 225, p < 0.0001). The NREM durations and REM durations at 7 and 10 h TIB differed significantly (p < 0.0001 for all) from those at 8.5 h TIB. The effect of TIB reduction on NREM duration resulted from a significant (F_2,149 = 545, p < 0.0001) stage N2 decrease. Stage N3 duration showed a trend (F_2,149 = 2.53, p = 0.084) to increase with TIB reduction.

Table 1.

Mean (± standard error) sleep stage data for the three time in bed conditions

	7 h	8.5 h	10 h
Total sleep time (min)	406 ± 1	472 ± 2	530 ± 2
NREM (min)	318 ± 2	360 ± 2	402 ± 2
N2 (min)	203 ± 2	250 ± 2	291 ± 3
N3 (min)	114 ± 2	111 ± 2	111 ± 2
REM (min)	89 ± 1	112 ± 1	128 ± 2
Sleep onset latency (min)	3.7 ± 1.3	11.2 ± 0.7	21.4 ± 1.4

We also observed significant reductions in TST min and NREM min as age increased. TST duration decreased (F_1,464 = 5.90, p = 0.016) by 2.0 min per year, and NREM duration decreased (F_1,464 = 7.53, p = 0.0063) by 1.7 min per year. REM duration did not change significantly with age (F_1,464 = 1.72, p = 0.19). The TIB effect on TST did not change significantly with age (TIB × age interaction, F_2,464 = 1.92, p = 0.15).

Effect of TIB on EEG power spectra

For the first 5 h of NREM (Figure 1A), FFT power in frequencies 1.1–5.4 Hz increased significantly with decreasing TIB. A more substantial change in the NREM spectrum was a significant decrease in power with decreasing TIB for frequencies between 11 and 14.2 Hz. Power was also decreased for frequencies between 7.9 and 8.7 Hz and between 9.7 and 9.9 Hz. For the EEG in REM cycles 2 and 3 (Figure 1B), we observed a significant increase in power with decreasing TIB for frequencies between 0.3–4.4 Hz and 5.0–5.2 Hz. For frequencies between 7.5 and 10.5 Hz, reducing TIB significantly decreased power. The main difference between the TIB reduction effects on the NREM and REM power spectra was a decrease in power in NREM EEG frequencies between 11 and 14.2 Hz, which was not seen in REM sleep. To test whether this difference simply resulted from examining different portions of the night, we examined TIB effects on the EEG power spectrum in NREMPs 2 & 3 (Figure 1C). These TIB effects were similar to those found for the first 5 h of NREM sleep except that the range of frequencies showing a significant increase was limited to 1–4 Hz and the range of frequencies showing a significant decrease in power with reduced TIB extended up to 20 Hz.

Figure 1.

EEG power spectra with average power at 7 h TIB expressed as a percentage of average power at 10 h TIB for (A) first 5 h of NREM, (B) REM periods 2 and 3, and (C) NREM periods 2 and 3. Bars on the x-axis indicate frequencies for which power changed significantly (α = 0.01) with TIB. Decreasing TIB increased EEG low frequency power and reduced EEG alpha power in both NREM and REM sleep. Decreasing TIB reduced EEG sigma power in NREM sleep but not REM sleep. Although the TIB data were averaged over all 3 years of the study for this figure, age was included as a factor in the statistical analyses which examined effects on log power rather than power.

Effect of TIB on the delta (1–4 Hz) EEG in NREM sleep

All-night delta energy was slightly (5.5%) smaller for 7 h TIB than for 10 h TIB (Figure 2A), but the decrease in delta energy with decreasing TIB was not significant (F_2,149 = 2.67, p = 0.073). The decrease in NREM duration far exceeded the small nonsignificant decrease in delta energy because slow-wave EEG activity is concentrated at the start of the night. As a result, all-night delta power (energy/time) increased significantly (F_2,149 = 43.5, p < 0.0001) with decreasing TIB.

Figure 2.

Average (± 1 standard error) delta (1–4 Hz) energy or power for each of the three TIB schedules. (A) All night NREM delta energy showed a trend (p = 0.073) towards declining with decreasing TIB. (B) Decreasing TIB was associated with a small but significant increase in delta power in the first 5 h of NREM sleep. Energy and power were not normally distributed; therefore, statistical analyses tested TIB (and age) effects on log energy and log power.

To eliminate the effects of variations in NREM duration, we examined the effect of TIB on delta in the first 5 h of NREM sleep (Figure 2B), an NREM duration common to all three TIB conditions. Decreasing TIB significantly increased (F_2,148 = 7.31, p = 0.0009) power in the first 5 h of NREM sleep. The TIB effect was not linear; delta power was significantly (p = 0.004) greater for 7 h TIB than for 8.5 h TIB which did not differ (p = 0.53) from delta power for 10 h TIB. Average delta power for 7 h TIB was 4.4% greater than for 10 h TIB. As age increased, there was an expected significant decrease (F_1,445 = 351, p < 0.0001) in delta power. The TIB effect did not change with age (TIB × age interaction, F_2,148 = 0.65, p = 0.52). Confirming that 5 h of NREM sleep was a duration common to the three TIB conditions, NREM duration did not change significantly with TIB (F_2,148 = 0.74, p = 0.48). However, varying TIB durations did alter the composition of the first 5 h of NREM sleep. Stage N2 duration significantly decreased (F_2,148 = 16.0, p < 0.0001) with decreasing TIB, and stage N3 significantly increased (F_2,148 = 15.8, p < 0.0001) with decreasing TIB.

Effect of TIB and age on the distribution of delta throughout the night

The analyses below included only nights with four complete NREM periods on night 4. TIB did not significantly affect (F_2,144 = 0.09, p = 0.92) the percentage of all-night delta energy in the first NREMP, nor did age (F_1,423 = 2.03, p = 0.15). We also examined the effects of TIB and age on the parameters of the Process S equation that describes the exponential decline in delta across the night. Neither TIB, treated as a continuous measure for this analysis, nor age had a significant effect (p > 0.2 for all) on any of the process S parameters.

Effect of TIB on sigma (11–15 Hz) power in NREM EEG

As shown in Figure 3A, for all-night NREM EEG, reducing TIB decreased sigma energy (F_2,148 = 140, p < 0.0001), with significant decreases in energy as TIB was reduced from 10 to 8.5 h (p < 0.0001) and from 8.5 to 7 h (p < 0.0001). From 10 h TIB to 7 h TIB, sigma energy decreased from 0.504 to 0.361 mV² s, a 39.6% decrease. Decreasing TIB significantly reduced (F_2.148 = 19.5, p < 0.0001) sigma power in the first 5 h of NREM (Figure 3B). Reducing TIB from 10 to 8.5 h reduced sigma power (p = 0.0059) as did reducing TIB from 8.5 to 7 h (p = 0.0009). From 10 h TIB to 7 h TIB, sigma power decreased from 21.8 to 19.4 µV², a 12.2% decrease. As age increased, sigma power decreased significantly (F_1,454 = 52.9, p < 0.0001). The age by TIB interaction did not show a change in the TIB effect with age (F_2,454 = 1.91, p = 0.15).

Figure 3.

Average (± 1 standard error) sigma (11–15 Hz) energy or power for each of the three TIB schedules. Decreasing TIB greatly reduced all night sigma energy (A) and reduced average power in the first 5 h of NREM sleep (B). Energy and power were not normally distributed; therefore, statistical analyses tested TIB (and age) effects on log energy and log power.

The following results are for an analysis that included only nights with four complete NREM periods on night 4. Cycle and TIB were both treated as categorical factors in this analysis. Sigma power significantly differed (F_3,228 = 58.9, p < 0.0001) between cycles. As shown in Figure 4, there was a reduction in power from cycle 1 to cycle 2 followed by an increase in power across cycles 2 through 4. We also found a significant TIB by cycle interaction (F_6,432 = 3.15, p = 0.0050). Cycle 2 showed the largest reduction in sigma power for the 7 h compared to the 10 h TIB condition and sigma power in cycle 4 was the least affected by TIB. We found no significant age by cycle interaction (F_3,1920 = 1.32, p = 0.26) and no significant age by TIB by cycle interaction (F_6,1920 = 0.13, p = 0.99).

Figure 4.

The trend of sigma (11–15 Hz) power across NREMPs 1–4 for each of the three TIB schedules. TIB effects differed by cycle with the greatest TIB effect on sigma power in NREMP 2 and the smallest effect on sigma power in NREMP 4. Energy and power were not normally distributed; therefore, statistical analyses tested TIB (and age) effects on log energy and log power.

Discussion

Current sleep theory is guided by two general assumptions. First, that sleep serves some functions that are restorative for waking brain physiology. Second, these restorative functions include active neurometabolic processes as well as what appears to be the passive restoration of metabolic substrates due to depressed neural activity. The application of our simple TIB dose–response paradigm in this longitudinal study of adolescence, a period of rapid age-related changes in brain and sleep electrophysiology, has enabled us to make observations relevant to both these general questions. Thus, in a recent report [24] from this study, we showed that the modest TIB reduction from 10 to 7 h reduced subsequent waking EEG power in a wide range of frequencies with the effect being largest in the alpha frequencies. The main findings we report here are concerned with sleep EEG. They include: (1) TIB reduction reduced both NREM and REM sleep duration, (2) an increase in NREM and REM delta power with sleep restrictions (which contradicts our initial report [15]), (3) a decrease in alpha power in both NREM and REM sleep with sleep restriction, and (4) a selective reduction in NREM sigma power and energy with sleep restriction.

Effects of time in bed and age on EEG-measured sleep durations

Reducing TIB effectively reduced total sleep durations. Results found here for adolescents across ages 9.9–16.2 years, are similar to the findings of the first year of this study [15] in that reducing TIB reduces both REM and NREM sleep durations. This reduction of both NREM and REM sleep differs strikingly from the selective reduction of NREM duration across ages 9–18 years that we found in an earlier longitudinal study of 67 subjects sleeping on their self-selected weeknight sleep schedules [8]. This selective age-related reduction in NREM sleep duration argues that the brain changes of adolescence selectively reduce the biological need/capacity to generate NREM sleep. This interpretation is consistent with our original hypothesis that the brain’s recuperation for the plastic neural activities of waking primarily takes place in NREM sleep [11]. According to this model, waking neural activities become less intense across adolescence because of late synaptic elimination and the consequent declines of waking brain metabolic rate [25]. An alternative model proposes that a circadian phase delay [26, 27] pushes bedtimes to a later hour. The model proposes that this delay combined with early rise times reduces sleep duration across adolescence despite a persistent sleep need. This phase delay model does not explain the selective age-related decline in NREM sleep. Our ongoing studies, which will extend our sample to age 22 years, along with more detailed analysis of the present data set, may enable us to distinguish between these alternative explanations of the decline in sleep duration across adolescence.

TIB effects on delta EEG energy and power

Although delta energy was slightly (5.5%) smaller for the 7 h than the 10 h TIB condition, this effect did not reach statistical significance. In other words, the total amount of slow-wave EEG, delta, activity accumulated in a night of sleep was not significantly affected by reducing TIB by 3 h. This result can be explained by the typical distribution of delta across the night. Slow-wave EEG activity is normally concentrated in the first part of the night. The last 3 h in bed includes very little stage N3 sleep; therefore, eliminating these 3 h had little effect on the total delta activity of the night.

To determine if reducing TIB alters the need for the recuperative processes, we examined the effect on average delta power in the first 5 h of NREM sleep. Delta power increased slightly but significantly with reduced TIB. Note that reducing TIB from 10 to 8.5 h did not significantly affect delta power. Reducing TIB to 7 h increased delta power either because wake duration increased or because sleep duration decreased to an extent that it produced a small increase in the need for the recuperation provided by slow-wave EEG activity. Delta power in REM sleep as well as in NREM sleep was significantly increased. These findings are consistent with the effects of total sleep deprivation which also increases delta power in both NREM and REM sleep [28]. The NREM and REM delta increases found here contradict our previous report, based on the first year of data from this study, which found that TIB reduction did not significantly increase delta power [15]. Our current delta findings are also inconsistent with the findings with more substantial sleep restriction in young adults [12, 13]. These studies show that even sleep restriction to 4 h/night for 14 consecutive days does not produce an increase in delta [14]. In acute experiments, a robust NREM delta response to partial deprivation seems to require the loss of delta from the first or second NREM periods [29].

Thus, our current findings in adolescents differ from previous results in young adults. This difference might be due to the greater statistical power of the current study (with 527 nights of data from the 77 participants) which allowed us to detect a small increase (4.4%) in delta power. A more interesting possibility is that adolescents are more sensitive to sleep loss than are young adults. Some evidence supporting this possibility was the delta power increase that Ong et al. [30] found in adolescents who were limited to 5 h in bed for 5 nights. However, if children’s brains are more susceptible to sleep reduction, we would expect the magnitude of the delta power effect to decrease with age. In the present study, age and TIB effects on delta power did not interact over the 10- to 16-year age range studied. Further research over a wider age range may help distinguish between these and other possibilities. The absence of a TIB effect on the decreasing trajectory of delta across the night is further evidence that 3 h TIB restriction produces only a small perturbation of the recuperative processes of slow-wave EEG.

TIB effects on alpha power in NREM and REM sleep

In addition to producing small increases in NREM and REM delta power, reducing sleep from 10 to 7 h in bed produced decreases in alpha power in both NREM and REM sleep. Taken in association with our previous finding that this amount of sleep reduction also reduces alpha power in waking EEG [24], it shows that insufficient sleep reduces alpha EEG power in all three qualitatively different states of human brain organization: waking, NREM, and REM sleep. Although alpha waves have been intensively studied since Berger’s discovery of the EEG, their significance for brain function still has not been established. A close association of alpha waves to visual function has long been indicated by the strong increase of alpha power with eye closing but it is only in recent years that it was discovered that this eye-closing response is not limited to EEG in the primary visual cortex but is manifest in many if not most brain areas [31, 32]. We have speculated that this wide-spread response may indicate the overriding importance of visual information for the human brain [24].

Sleep restriction reduces sigma energy and power in NREM sleep

Another unexpected result of this study was that relatively modest sleep restriction strongly reduced both all-night sigma energy and sigma power within NREM sleep. In other words, the 23% reduction in TST with TIB restriction from 10 to 7 h was associated with not only a 40% loss of all-night accumulated sigma activity but also with a significant decrease in the rate of sigma production. Uchida et al. [33] suggested that the bulk of power in NREM sigma frequencies (12–15 Hz) is contributed by organized sleep spindles, which are of distinctly higher amplitude than most of the background waves of these frequencies. Uchida’s conjecture subsequently received some empirical confirmation in a study by Dijk et al. [34]. The TIB effect on sigma frequency in the current study was specific to NREM sleep, a result also consistent with a depression of sleep spindle activity. Reynolds et al. [35], measured the effect of sleep restriction on sleep spindle characteristics and found that 5 h TIB for five consecutive nights reduced amplitude of fast spindles. Contrary to the large sigma activity reduction found with the 7 h TIB condition in our study, Reynolds et al. did not find a spindle difference between participants keeping 7.5 and 10 h TIB schedules. The contradiction may be related to the older age, 15–17 years, of their participants. Under the interpretation that sigma EEG reflects spindle activity, the depressed power in 11–15 Hz with sleep restriction could implicate thalamocortical networks, which appear to control both delta waves and sleep spindles [36]. Considering the inverse relationship between power in sigma and delta frequencies [33], it is not surprising that an increase in delta power with reduced TIB should be accompanied by decreased sigma power. The larger size of the effect and the altered pattern of sigma power across the night would argue that the TIB reduction of sigma power is likely to be of greater biological significance than the small delta increase. A post-hoc analysis of differences in night 2 and night 4 sigma activity was used to determine if the sigma effect increased across the period of TIB restriction. Neither all night energy (F_1,76 = 0.02, p = 0.87) nor power (F_1,76 = 0.00, p = 0.99) in the first 5 h or NREM differed between nights, nor did energy (F_2,144 = 0.08, p = 0.93) or power (F_2,144 = 0.08, p = 0.93) show a TIB by night interaction. We had expected that the strong TIB effect might accumulate across the multiple nights of restricted sleep, as does daytime sleepiness and performance [37]. A longer study with nightly recordings might provide more conclusive evidence regarding cumulative effects on sigma EEG.

The strong TIB effect on sigma energy and power is pertinent to several current research questions. Sleep spindles have been proposed as markers of sleep-dependent memory processes, and/or general mental ability [38, 39]. Does reducing spindle activity via sleep restriction impair putative memory processing during sleep? Does it impair cognitive function during waking? If the answers are “yes,” the large reduction in sigma energy with 3 h of TIB reduction may explain how sleep loss impairs daytime performance. Furthermore, the impacts of sleep loss might be particularly strong in childhood and adolescence. However, for sigma as for delta, the TIB effect on energy and power did not change with age. We note, also, that the evidence for a relation between spindles and memory is not strong in children [40]. The strong NREM specific effect of TIB restriction on sigma activity found in the current study indicates that our simple TIB manipulation might provide a way to test the relation of sleep spindles to memory and other cognitive functions.

Age effects on delta, alpha, and sigma EEG power

EEG power in the delta, alpha, and sigma frequency bands decreased significantly across the 10–16 years age range of this study. As we have previously proposed [3, 25], we attribute at least part of this decrease to synaptic pruning during adolescence [41]. EEG is a recording of synchronous activity of large pools of cortical neurons. As neuronal connections are eliminated the number of neurons oscillating in unison decreases and EEG power declines. The resulting changes in the brain might have altered the response to the sleep loss associated with the TIB restriction of this study; however, the absence of a significant TIB by age interaction on power in any frequency bands indicates otherwise.

When only night 4 data were analyzed, our analysis did not show a significant age effect on the distribution of delta across the night. The absence of an age effect contradicts what we previously found in our larger longitudinal study [23]. To determine if increasing the sample size would affect the outcome of the analyses, we ran post-hoc analyses using data from both nights 2 and 4. NREMP1 percent delta energy decreased significantly with age (F_1,909 = 8.33, p = 0.0040) by 0.59% per year from an intercept of 50.2% at age 13.2. The age effect on the parameters of the equation that describes the exponential decline in delta across the night remained nonsignificant (p > 0.2 for all parameters). Despite this non-significant result, we are confident in the findings of our prior study which included more recordings and recorded individual subjects semiannually for up to 7 years.

Limitations

The current study includes participants from ages 9.8 to 16.2 years of age. A few TIB by age interactions approached the 0.05 level of significance. Expanding the age range might allow us to better determine whether these effects of sleep restriction change with age. In other reports from this study, we have pointed to potential limitations related to subject recruitment, enrollment, and retention. Thus, only participants who could tolerate sleep restriction to 7 h in bed for four consecutive nights participated in this study. In fact, several younger participants withdrew during the first year because they could not tolerate the discomfort produced by restriction to 7 or even 8.5 h in bed. It is possible that, had we been able to retain these subjects, we would have found even larger effects of sleep restriction. Furthermore, allowing participants’ input into the scheduling of the different TIB conditions rather than randomizing the assignments may have contributed to the EEG differences that we recorded. Our method of shortening TIB by delaying bedtime was an attempt to mimic the TIB shortening that occurs across adolescence. Although we have attributed the EEG effects to the TIB changes and resultant sleep duration changes, some of the EEG effects may result from circadian shifts.

Conclusion

As predicted by homeostatic models, TIB restriction significantly increased delta EEG power in both NREM and REM sleep although the effects were small. TIB restriction more substantially reduced alpha power in both NREM and REM sleep as well as (as we recently reported [24]) in waking EEG. Another novel finding was a marked reduction in NREM sigma energy and power with sleep restriction. Although speculative, we interpret the increase in delta power following sleep restriction as an increase in the passive restoration processes of sleep (as evidenced by decreased brain metabolic rate) and the decreases in alpha and sigma power as evidence of impairment of hypothesized active restorative processes of sleep. Elucidating their physiologic and metabolic mechanisms could help unravel the functions of sleep and the specific roles played by these EEG frequencies. The statistically reliable effects of modest sleep restriction found here suggests that our paradigm could be a generally useful tool for further sleep research.

Disclosure Statement

Financial disclosure: none.

Non-financial disclosure: none.