The effect of sleep-specific brain activity versus reduced stimulus interference on declarative memory consolidation


Correspondence Christoph Nissen, MD, Department of Psychiatry and Psychotherapy, University Medical Center Freiburg, Hauptstr. 5, 79104 Freiburg, Germany. Tel.: +49(0)761-270-65010; fax: +49(0)761-270-66190; e-mail:


Studies suggest that the consolidation of newly acquired memories and underlying long-term synaptic plasticity might represent a major function of sleep. In a combined repeated-measures and parallel-group sleep laboratory study (active waking versus sleep, passive waking versus sleep), we provide evidence that brief periods of daytime sleep (42.1 ± 8.9 min of non-rapid eye movement sleep) in healthy adolescents (16 years old, all female), compared with equal periods of waking, promote the consolidation of declarative memory (word-pairs) in participants with high power in the electroencephalographic sleep spindle (sigma) frequency range. This observation supports the notion that sleep-specific brain activity when reaching a critical dose, beyond a mere reduction of interference, promotes synaptic plasticity in a hippocampal-neocortical network that underlies the consolidation of declarative memory.


Sleep after learning has been shown to facilitate the consolidation of new and initially unstable memories and underlying long-term synaptic plasticity in comparison to equal periods of active wakefulness (Diekelmann and Born, 2010). Significant benefits on memory have been observed after both 8 h of night-time sleep and short midday naps of 1–2 h (Mednick et al., 2003). However, fundamental questions persist whether sleep-specific brain activity actively strengthens new memories (active consolidation hypothesis), or whether the interruption of novel sensory input and motor activity co-occurring with sleep just protects the novel memories from disruptive interference in a critical period after acquisition (interference hypothesis).

The active consolidation hypothesis is supported by animal (Wilson and McNaughton, 1994) and human studies (Rasch et al., 2007) showing a task-related ‘off-line’ replay of neuronal activity during sleep after training. Further evidence suggests consolidation effects of highly synchronized oscillatory patterns in a thalamo-cortical network during non-rapid eye movement (NREM) sleep, including sleep spindles (Clemens et al., 2005) and electroencephalographic (EEG) slow-wave activity (Marshall et al., 2006). More specifically, sleep spindles, brief but powerful bursts of synchronous neuronal firing during mammalian stage 2 sleep in the EEG sigma (12–16 Hz) frequency range, might promote sleep-related synaptic plasticity underlying memory consolidation by synaptic strengthening (Rosanova and Ulrich, 2005). In contrast, slow-wave activity (0.1–3.5 Hz) and underlying ‘down-states’ of hyperpolarization with neuronal silencing and subsequent ‘up-states’ of depolarization with intense neuronal firing (Steriade, 2006) might sharpen the information-to-noise ratio and new memory representations through synaptic downscaling (Tononi and Cirelli, 2006).

The interference hypothesis is supported by studies demonstrating enhanced memory retrieval due to protection of new memory traces from disrupting input during a critical period after acquisition, independently of sleep (e.g. Gottselig et al., 2004; Mednick et al., 2009). More recently, researchers assume that during periods of quiet rest, just as during sleep, reduced interference and reduced encoding of new memories might facilitate the evolution of memory-consolidating processes in the brain (opportunistic consolidation hypothesis; Mednick et al., 2011). Thus, for instance, neuronal replay has been observed during both sleep and restful waking with reduced interference (Foster and Wilson, 2006).

To date, only a few studies have directly compared the effect of sleep versus reduced interference on memory (Gottselig et al., 2004; Mednick et al., 2002, 2009). These studies employed daytime sleep (nap) paradigms to minimize the confounding effects of sleep deprivation, stress and circadian bias inherent to whole-night studies. Although there is convincing evidence that sleep yields beneficial effects on memory consolidation, to date none of these studies has directly demonstrated that sleep after learning is superior to quiet wakefulness with attenuated interference.

The current study tested the hypothesis that sleep-specific brain activity, specifically EEG sigma power as an index of sleep spindle activity (Dijk et al., 1993) and EEG slow-wave activity (Tononi and Cirelli, 2006), rather than a mere reduction of interference, enhances the consolidation of novel declarative memories in adolescents – a period of particular importance for learning, memory and plastic changes of the brain.

Materials and methods


A total of 85 subjects were recruited from secondary schools in the Freiburg area. Fourteen subjects were excluded after initial screening, 22 subjects did not meet the sleep/wake criteria for inclusion as defined below; = 49 were included in the present analysis. The study had been approved by the local ethic committee, and was carried out according to the Declaration of Helsinki. Written informed consent was obtained from all participants and their parents (or legal representatives). To reduce variance related to sex, age and the level of education, the sample consisted of female adolescents from secondary high schools aged 16 years. All subjects were post-pubertal and tested in the follicular phase of their menstrual cycle. Their general IQ level was estimated at 106.6 ± 10.1 based on the Standard Progressive Matrices (Heller et al., 1998). All subjects were free of any relevant physical or mental disorder, sleep disorder, and medication or substance use. All participants followed a regular sleep/wake pattern, with an averaged sleep duration of about 8 h per night within the week prior to and during the study as determined by sleep diaries and actigraphy. Overall, the selection of participants ensured the investigation of a healthy and homogenous adolescent sample with a comparable developmental and cognitive status, and well-regulated day-to-day routines.

Study design

All subjects (= 49) underwent a memory and sleep/wake protocol with polysomnographic monitoring from 13:30 hours to 14:30 hours (Fig. 1, split-plot design). They were adapted to the sleep laboratory condition prior to the study. Subjects in Group 1 (= 25) underwent a sleep and an active waking condition (video, physical activity). Subjects in Group 2 (= 24) underwent a sleep and a passive waking condition (maximally reduced interference). The sleep and wake conditions within both groups were separated by 1 week and implemented in a counterbalanced order to control for sequence effects.

Figure 1.

Study design. After adaptation to the sleep laboratory (not shown), all subjects (= 49) were assigned to a sleep condition with polysomnographic monitoring from 13:30 hours to 14:30 hours, with half of them being additionally allocated to an active waking condition (Group 1, = 25), and half of them to a passive waking condition (Group 2, = 24). The sleep/wake conditions were separated by 1 week and implemented in a counterbalanced order. Memory and general cognitive performance were assessed in the morning prior to and in the afternoon after the sleep/wake intervention.

Memory and general cognitive performance were assessed prior to (11:00–13:00 hours) and after (15:00–16:00 hours) the sleep/wake intervention (13:30–14:30 hours) using a declarative word-pair task and a standardized neuropsychological test battery. An interval of 30 min between the end of the sleep/wake condition and the retrieval session was implemented in order to minimize potential effects of sleep inertia. Cortisol saliva specimens were collected half-hourly before (10:30–13:30 hours) and after (14:30–16:00 hours) the sleep/wake intervention to control for potential effects of stress.

Experimental conditions

During the sleep condition, subjects were instructed to sleep while lying in bed in the dark and quiet sleep laboratory with polysomnographic monitoring.

During the active waking condition, subjects were watching a video and playing table tennis to ensure an elevated, but controlled level of sensory input and motor activity. The condition was continuously monitored by trained sleep laboratory staff.

During the passive waking condition, subjects were instructed to lie awake in bed in the dark and quiet sleep laboratory with maximally reduced sensory input and motor activity (interference). Polysomnography was monitored online. Discrete wake signals (tones via intercom) were applied at the first signs of sleep onset within the first 30 s of each epoch. To ensure clear-cut differences between the conditions for primary statistical comparisons, inclusion criteria were set at ≥ 50% sleep efficiency (sleep condition) and ≤ 10 wake signals (passive waking condition).

Sleep recordings

Polysomnography was recorded during the adaptation, sleep and passive waking condition for 1 h from ‘lights off’ (13:30 hours) until ‘lights on’ (14:30 hours). All recordings included an EEG (C3-A2, C4-A1), horizontal and vertical electrooculogram, submental electromyogram and an electrocardiogram. Polysomnographic recordings were visually scored off-line by experienced raters according to standard criteria (Rechtschaffen and Kales, 1968). The following variables of sleep continuity and architecture were evaluated: time in bed (period between lights off and lights on); total sleep time [time spent in sleep stage 2, slow-wave sleep (SWS) and rapid eye movement (REM) sleep]; sleep latency (period between lights off and first occurrence of sleep stage 2, SWS or REM sleep); sleep efficiency (ratio of total sleep time to time in bed × 100%); as well as the time spent in waking and in sleep stages 1 and 2, SWS and REM sleep.

Additionally, EEG spectral analysis was performed on the C3-A2 derivation in 30-s epochs for which sleep stages had been determined. Data were recorded with a sampling rate of 200 Hz and a resolution of 16 bits. Signals were recorded with a time constant of 0.3 s and low-pass-filtered at 70 Hz. Spectral estimates for each epoch were obtained as the average of 22 overlapping FFT windows (512 data points, 2.56 s) covering a 30-s epoch to obtain the spectral power within that epoch, resulting in a spectral resolution of 0.39 Hz. A Welch taper was applied to each FFT window after demeaning and detrending the data in that window. The spectral power values were then log-transformed (base e) and continuously stored on disk. All subsequent steps including statistical analysis were performed on these logarithmic values, which have more symmetrically distributed errors than raw spectral power. Artefact rejection used an automatic method discarding epochs due to abnormal total or gamma-band power values relative to a 10-min moving window. The log spectra of the remaining epochs were averaged across all NREM sleep epochs. Spectral band power was calculated for the following theoretical frequency ranges: delta 0.1–3.5 Hz (delta1 0.1–1.5 Hz; delta2 1.5–3.5 Hz); theta 3.5–8 Hz; alpha 8–12 Hz; sigma 12–16 Hz (sigma1 12–14 Hz; sigma2 14–16 Hz); beta 16–24 Hz (beta1 16–20 Hz; beta2 20–24 Hz); and gamma 24–50 Hz.

Declarative memory (word-pair task)

In the word-pair task, 46 semantically related word-pairs (e.g. bird–eagle) were presented randomly on a 15″ computer screen for 5000 ms, followed by a 100-ms blank screen using the Presentation® software (word-pair list and procedures from Marshall et al., 2006). Four additional word-pairs at the beginning and at the end of the task served to buffer primacy and recency effects. During the learning session in the morning (baseline), the word-pairs were presented repeatedly in a randomized order until the participants remembered at least 60% in a cued recall test, i.e. presenting the first word of the previously learned word-pairs (correct matches were not displayed after the subjects' responses). During the retrieval session in the afternoon, participants performed a cued recall test without prior presentation of the word-pairs. Memory encoding was assessed as the number of correctly retrieved words in the last learning trial in the morning session. Memory consolidation was calculated as the percentage of correctly retrieved words in the afternoon referred to the number of correctly encoded words in the morning session (retention rate,%). Parallel versions of the task were used for repeated measurements.

Control variables

To control for possibly confounding effects, parameters of general cognitive performance and stress were assessed. Cognitive performance, including alertness (Test for Attentional Performance; Zimmermann and Fimm, 2007) and working memory (Digit Span and Block Tapping; Tewes, 1994), was investigated prior to the morning and afternoon test sessions.

Furthermore, cortisol saliva specimens were collected on the test days half-hourly before (10:30–13:30 hours) and after (14:30–16:00 hours) the sleep/wake intervention. The resulting 20 samples per subject (10 per block) were analysed by standard procedures (Voderholzer et al., 2011).

Data analysis

Data were analysed using the Statistical Package for Social Sciences (spss) version 18.0. Means and standard deviations were calculated for descriptive purposes. Analyses of variance (anova) were used to test for baseline differences in demographic characteristics and general cognitive performance between the experimental groups. Repeated-measures 2 × 2 anova with the within-subject factor Condition (sleep versus wake) and the between-subject factor Group [Group 1 (sleep/active waking) versus Group 2 (sleep/passive waking)] were calculated to compare sleep variables, and parameters of memory encoding and memory consolidation. For comparison of memory consolidation in the stratified analysis, a 2 × 2 repeated-measures anova with the within-subject factor Condition (sleep versus wake) and the between-subject factor Group (high versus low spindle activity) was conducted. Post hoc comparisons were performed using single anovas. The impact of cortisol levels (AUC; area under the curve) was investigated by a 2 × 2 repeated-measures anova with the within-subject factor Condition (sleep versus wake) and Time-of-day (morning versus afternoon). Bivariate Pearson correlation analyses were used to analyse the relationships between IQ, memory encoding, memory consolidation and sleep variables. The level of significance was set at P ≤ 0.05 (two-tailed).


Sleep parameters

Statistical analyses revealed highly significant differences in the sleep parameters between the sleep and wake conditions (Table 1; = 3097.4, < 0.001). These clear-cut differences demonstrate the feasibility and effectiveness of the study protocol. The chosen sleep lab interval of 1 h ensured that participants showed virtually no periods of REM sleep (only five subjects showed brief periods of REM sleep), and allowed for the comparison of well-defined periods of NREM sleep with periods of active and passive wakefulness. Of note, the exclusion of subjects with brief periods of REM sleep did not alter the results described in the following sections.

Table 1. Sleep parameters for the total sample (= 49)
  Group 1 (active) N = 25 Group 2 (passive) N = 24
  1. Sleep stages are given in min, and as a percentage of time in bed.

  2. EEG, electroencephalogram; REM, rapid eye movement; SWS, slow-wave sleep.

Sleep continuity
Time in bed, min0.0 ± 0.060.4 ± 1.457.4 ± 12.360.5 ± 1.2
Total sleep time, min0.0 ± 0.043.1 ± 8.60.2 ± 0.641.0 ± 9.1
Sleep efficiency (%)0.0 ± 0.076.3 ± 11.91.1 ± 2.974.0 ± 11.0
Sleep latency, min0.0 ± 0.011.0 ± 5.84.8 ± 10.011.9 ± 5.1
No. of awakenings0.0 ± 0.00.0 ± 0.01.7 ± 2.60.0 ± 0.0
Sleep architecture
Waking, min0.0 ± 0.010.5 ± 6.755.0 ± 12.511.7 ± 5.8
Waking (%)0.0 ± 0.017.3 ± 10.991.7 ± 20.819.3 ± 9.6
Stage 1, min0.0 ± 0.06.8 ± 4.12.0 ± 3.37.6 ± 4.5
Stage 1 (%)0.0 ± 0.011.2 ± 6.83.3 ± 5.612.7 ± 7.5
Stage 2, min0.0 ± 0.023.9 ± 7.10.2 ± 0.624.4 ± 8.0
Stage 2 (%)0.0 ± 0.039.6 ± 11.80.4 ± 1.140.5 ± 13.5
SWS, min0.0 ± 0.018.2 ± 10.60.0 ± 0.016.3 ± 12.8
SWS (%)0.0 ± 0.030.2 ± 17.50.0 ± 0.026.8 ± 21.3
REM sleep, min0.0 ± 0.01.0 ± 3.00.0 ± 0.00.3 ± 1.0
REM sleep (%)0.0 ± 0.01.6 ± 5.00.0 ± 0.00.5 ± 1.6
EEG spectral power (log)
Delta 10.0 ± 0.05.6 ± 0.50.0 ± 0.05.5 ± 0.6
Delta 20.0 ± 0.05.3 ± 0.40.0 ± 0.05.2 ± 0.5
Theta0.0 ± 0.04.1 ± 0.40.0 ± 0.04.1 ± 0.4
Alpha0.0 ± 0.02.9 ± 0.40.0 ± 0.02.8 ± 0.4
Sigma0.0 ± 0.02.5 ± 0.30.0 ± 0.02.5 ± 0.5
Sigma stage 2 sleep0.0 ± 0.02.7 ± 0.30.0 ± 0.02.7 ± 0.4
Beta 10.0 ± 0.0−1.9 ± 0.30.0 ± 0.0−1.8 ± 0.4
Beta 20.0 ± 0.02.1 ± 0.30.0 ± 0.02.1 ± 0.5
Gamma0.0 ± 0.01.3 ± 0.40.0 ± 0.01.3 ± 0.6

Declarative memory consolidation

Unstratified analysis

When analysing the total sample (= 49) at baseline, the groups (active/passive) and conditions (sleep/wake) did not differ in the number of trials to criterion or in the final number of correctly encoded word-pairs (Table 2; Group: = 0.6, = 0.648; Condition: = 1.5, = 0.241; Group × Condition: = 0.7, = 0.575). The number of learning trials to criterion was not related to post-sleep/wake memory consolidation (sleep: = 0.222, = 0.125; waking: = 0.104, = 0.475).

Table 2. Declarative memory performance for the total sample (= 49) in the morning before and in the afternoon after the sleep and active/passive waking conditions
  Group 1 (active) N = 25 Group 2 (passive) N = 24
  1. Encoding: number of correctly recalled word-pairs in the morning session. Retrieval: number of correctly recalled word-pairs in the afternoon session.

No. of trials to criterion1.3 ± 0.51.2 ± 0.41.4 ± 0.51.4 ± 0.6
Encoding36.8 ± 4.735.6 ± 5.136.2 ± 4.936.9 ± 4.0
Retrieval35.9 ± 5.235.1 ± 5.635.6 ± 5.336.5 ± 4.5

As analysed in a first step, against our primary hypothesis, declarative memory consolidation from the morning to the afternoon test session (retention rate,%) did not differ significantly between the groups or conditions (Group: = 0.4, = 0.536; Condition: = 0.7, = 0.417; Group × Condition: < 0.1, = 0.944). Post hoc tests did not reveal a significant sleep/wake effect, neither for Group 1 (sleep/active waking: = 0.3, = 0.585) nor for Group 2 (sleep/passive waking: = 0.4, = 0.530). Fig. 2 visualizes the similar retention rates across conditions. On the basis of highly similar values in the sleep and waking conditions, participants of the active and the passive waking group were combined for all subsequent analyses.

Figure 2.

Word-pair retention rates from the morning to the afternoon test session (total sample, = 49) show no significant difference between the sleep and waking conditions. Error bars depict standard errors.

Correlation with EEG sigma (sleep spindle) activity

Consistent with our primary hypothesis, Pearson correlation analyses revealed a significant positive correlation between declarative memory consolidation (retention rate,%) and EEG power in the sleep spindle frequency range (sigma, 12–16 Hz) during NREM sleep across the sleep condition (= 49; = 0.288, = 0.045). The correlation between word-pair retention rate and EEG sigma power was strengthened when considering EEG sigma power specifically during stage 2 sleep, which is particularly specific for sleep spindle activity (= 0.393, = 0.005; Fig. 3). Splitting the sigma range during stage 2 sleep in slow and fast spindle activity yielded statistically significant correlations for the slow sleep spindle range (= 0.347, = 0.016), but not for the fast sleep spindle range (= 0.171, = 0.246). No other significant correlations between declarative memory consolidation and EEG spectral parameters during NREM sleep were observed. However, exploratory analyses revealed a negative correlation between memory consolidation and the duration of stage 2 sleep (Table 3).

Table 3. Correlations between declarative memory consolidation (word-pair retention rate,%) and sleep parameters (= 49)
 Word-pair retention rate (%)
r P
  1. Sleep stages are referred to in min, and as a percentage of the time in bed. Significant correlation effects are highlighted in bold.

  2. EEG, electroencephalogram; REM, rapid eye movement; SWS, slow-wave sleep.

Sleep continuity
Time in bed, min−0.0580.694
Total sleep time, min−0.1290.377
Sleep efficiency (%)−0.1190.415
Sleep latency, min0.1250.390
Sleep architecture
Waking, min0.1520.296
Waking (%)0.1570.283
Stage 1, min−0.1310.369
Stage 1 (%)−0.1260.388
Stage 2, min−0.337 0.018
Stage 2 (%)−0.328 0.021
SWS, min0.1430.327
SWS (%)0.1420.331
REM sleep, min0.1290.377
REM sleep (%)0.1300.372
EEG spectral power (log)
Delta 10.1720.238
Delta 20.2120.144
Sigma0.288 0.045
Sigma stage 2 sleep0.393 0.005
Beta 10.0910.536
Beta 20.1730.234
Figure 3.

Significant positive correlation between EEG power in the sleep spindle (sigma) frequency range during stage 2 sleep and word-pair retention rate across the sleep condition for the total sample (= 49; = 0.393, = 0.005).

To control for potential relationships between EEG sigma power and parameters of memory encoding or IQ, additional correlation analyses were performed. No significant correlations were observed between EEG sigma power and parameters of memory encoding (number of trials to criterion = 0.030, = 0.840, number of correctly recalled word-pairs in the last morning trial = −0.050, = 0.733) or general IQ (= 0.204, = 0.160), indicative of a consolidation-specific effect.

Analysis stratified by EEG sigma (sleep spindle) activity

In a second step and to follow-up on the observed correlation between EEG spectral power in the spindle (sigma) frequency range and declarative memory consolidation, we analysed the effect of sleep versus waking for subjects with high and low EEG sigma activity separately (median split). These subgroups did not differ with regard to IQ or other cognitive measures (> 0.1), baseline memory performance (> 0.7) or general sleep parameters, except for a reduced total sleep time and sleep efficiency in the subgroup with high EEG sigma activity during the sleep condition, due to slightly elevated waking times (Table 4). Of note are differences in EEG spectral parameters, showing higher values for almost all variables in the subgroup of participants with high EEG sigma activity, except EEG delta activity (Table 4).

Table 4. Sleep parameters for the subgroups with low and high EEG sigma activity (= 49; divided by median split), listed only for the sleep condition
  Low sigma activity N = 25 High sigma activity N = 24 Group effect
  1. Sleep stages are given in min, and as a percentage of time in bed. Significant effects are highlighted in bold.

  2. EEG, electroencephalogram; REM, rapid eye movement; SWS, slow-wave sleep.

Sleep continuity
Time in bed, min60.7 ± 1.560.1 ±
Total sleep time, min47.6 ± 5.943.1 ± 7.05.7 0.022
Sleep efficiency (%)78.4 ± 10.771.7 ± 11.64.4 0.041
Sleep latency, min10.3 ± 4.812.5 ±
Sleep architecture
Waking, min9.4 ± 5.812.9 ±
Waking (%)15.4 ± 9.421.4 ± 10.54.3 0.043
Stage 1, min6.2 ± 4.28.3 ±
Stage 1 (%)10.3 ± 6.913.7 ±
Stage 2, min24.9 ± 7.924.0 ±
Stage 2 (%)41.0 ± 13.240.1 ±
SWS, min19.6 ± 12.414.1 ±
SWS (%)32.4 ± 20.523.3 ±
REM sleep, min0.4 ± 1.10.9 ±
REM sleep (%)0.7 ± 1.91.5 ±
EEG spectral power (log)
Delta 15.6 ± 0.65.5 ±
Delta 25.2 ± 0.55.2 ±
Theta4.0 ± 0.44.2 ± 0.34.5 0.040
Alpha2.7 ± 0.43.0 ± 0.311.8 0.001
Sigma2.2 ± 0.22.8 ± 0.386.4 ≤ 0.001
Sigma stage 2 sleep2.4 ± 0.33.0 ± 0.250.1 ≤ 0.001
Beta 10.7 ± 0.31.1 ± 0.329.6 ≤ 0.001
Beta 2−0.6 ± 0.2−0.2 ± 0.329.9 ≤ 0.001
Gamma−2.0 ± 0.3−1.7 ± 0.316.9 ≤ 0.001

A subsequent anova revealed non-significant main effects on declarative memory consolidation for the factors Group (high versus low sigma; = 0.5, = 0.464) and Condition (sleep versus wake; = 0.8, = 0.375), but a significant Group × Condition interaction effect (= 4.1, = 0.050, medium to high effect size). Post hoc testing demonstrated a positive sleep effect on declarative memory consolidation in subjects with high EEG sigma activity (= 6.6, = 0.017; Fig. 4a). In contrast, this effect was not observed in subjects with low EEG sigma activity (= 0.5, = 0.498; Fig. 4b).

Figure 4.

(a) Word-pair retention rates for the stratified sample of participants with high EEG power in the sleep spindle (sigma) frequency range (= 24). Retention rates are calculated from the morning to the afternoon test session and show a significant sleep effect on declarative memory consolidation (= 6.6, = 0.017). Error bars depict standard errors. (b) Word-pair retention rates for the stratified sample of participants with low EEG power in the sleep spindle (sigma) frequency range (= 25). Retention rates are calculated from the morning to the afternoon test session and do not show a significant sleep effect on declarative memory consolidation (= 0.5, = 0.498). Error bars depict standard errors.

No impact of potentially confounding factors

Cognitive assessments prior to and after the sleep/wake conditions did not reveal performance differences between Groups or Conditions (Group: = 0.8, = 0.600; Condition: = 1.0, = 0.454; Group × Condition: = 0.9, = 0.512). This suggests that the effect observed for declarative memory consolidation cannot be explained by differences in alertness or working memory.

The analysis of saliva cortisol probes did not show significant differences in the morning or afternoon cortisol levels (AUC) between Conditions (< 0.1, = 0.958). We merely observed a significant time effect, indicating generally lower cortisol levels in the afternoon than in the morning (= 108.2, < 0.001). These results suggest that both waking conditions did not cause different levels of stress for the participants to the sleep condition.


The primary analysis of the present study did, against our initial hypothesis, not show a significant effect of sleep on declarative memory consolidation. However, in line with our hypothesis, we observed a significant positive correlation between sleep-specific brain activity in the form of EEG spectral power in the spindle (sigma) frequency range and declarative memory consolidation. Follow-up analyses revealed a beneficial effect of sleep over waking selectively under conditions of high EEG spindle (sigma) activity. This pattern of results is consistent with the active consolidation hypothesis of sleep, and provides indirect evidence against the interference hypothesis.

The finding of spindle-related memory effects is consistent with previous studies reporting a post-learning increase in the number, duration and density of spindles during night-time sleep (e.g. Gais and Born, 2004). In line with a recent study in young adults (Holz et al., 2012), selectively EEG power in the slow sleep spindle frequency range (12–14 Hz) was associated with overnight memory consolidation in the current study. With regard to potential mechanisms, sleep spindles have been reported to co-occur with hippocampal ripples that might promote a hippocampal-neocortical dialogue and synaptic refinements underlying declarative memory consolidation during sleep (Clemens et al., 2007). On a molecular level, oscillations in the spindle frequency range have been demonstrated to foster intracellular mechanisms that ultimately facilitate synaptic long-term potentiation thought to be the major molecular correlate of learning (Rosanova and Ulrich, 2005).

In contrast to some previous studies on night-time sleep, including work from our group (Holz et al., 2012), we did not observe a significant correlation between slow wave activity (SWA) and memory consolidation in the present study. This discrepancy might be explained by differences in the study designs. Specifically, the investigated period of sleep in the early afternoon of the current study is, according to the classical two-process model of sleep regulation (Borbély, 1982), associated with both: (1) an attenuated built-up of SWA (process S); and (2) a different circadian timing of the sleep period (process C), in comparison to night-time sleep. Yet a sufficient homeostatic pressure or the coincidence with a particular circadian phase might be a prerequisite for a profound effect of SWA on the sharpening of newly encoded memory traces, potentially mediated by synaptic downscaling (Tononi and Cirelli, 2006).

A number of limitations need to be addressed. First, it might be argued that subjects with high EEG spindle (sigma) power might be better sleepers or have higher IQs (Bódizs et al., 2005; Schabus et al., 2006). However, in our sample, the direct comparison between subjects with high and low sigma power did not reveal such relationships. Rather, our data showed a reduced total sleep time in participants with high EEG sigma activity, and – potentially due to a highly selected sample with a low variance in general cognitive abilities – no correlations with the IQ were observed. Second, the level of interference in our active waking condition might not have been high enough, and effects might only emerge under levels of highly similar sensory input and cognitive processing. Third, according to the standard model of declarative memory consolidation (Frankland and Bontempi, 2005), a novel memory for word-pairs strongly depends on the hippocampus within the first hours after encoding. The short retention interval of about 3 h in our study might be associated with similar recall performance on a behavioural level after the sleep and wake interventions, although sleep might have contributed to a hippocampal-cortical reorganization of the memory trace on a neural network level. Furthermore, note that our observations (in contrast to many studies on sleep and memory that have focused on male adults) are limited to female adolescents. All participants demonstrated post-pubertal status and were carefully controlled for their menstrual cycle. Previous studies suggest comparable effects of sleep on declarative memory consolidation in children, adolescents and adults, in contrast to conflicting findings for procedural memory (for review, please refer to Kopasz et al., 2010). However, future studies are needed to test whether our findings translate to male adolescents and subjects in other age ranges.

With regard to potential clinical implications, recent work suggests that the process of memory consolidation during sleep might be disrupted in patients with primary insomnia (Nissen et al., 2011) and obstructive sleep apnea (Kloepfer et al., 2009). Future work is needed to further investigate the alterations of the microstructure of sleep in patients with sleep disorders and their potential impact on memory. Finally, studies are needed to determine the effects of various treatments, including pharmacological and non-pharmacological interventions, on sleep and memory.

In conclusion, our finding that brief periods of NREM sleep with high EEG power in the spindle frequency range facilitate the consolidation of declarative memories is consistent with an active consolidation hypothesis of sleep and provide, at least indirect, evidence against a mere interference hypothesis. Future studies that manipulate distinct components of sleep (e.g. via transcranial direct current stimulation in humans or the induction of specific activity patterns in animals) are needed to further determine the cause–effect chain of the impact of sleep on memory.

Disclosure of potential conflicts of interest

Dr Voderholzer has received speaker honoraria from Sanofi-Aventis, Lundbeck, Pfizer, Cephalon and Lilly. He has been principal investigator of an investigator-initiated trial sponsored by Lundbeck. The other authors have declared no conflicts of interest.


The authors would like to thank the technical staff and the doctoral students Carla Minarik, Johannes Emrich and Stefanie Hoenle at the Department of Psychiatry and Psychotherapy, University Medical Center Freiburg, for their help in conducting the study. This study was supported by a research grant from the German Research Foundation to U.V. and C.N. (Vo 542/9-1).