Exploring the Effect of Sleep and Reduced Interference on Different Forms of Declarative Memory

Abstract

Study Objectives:

Many studies have found that sleep benefits declarative memory consolidation. However, fundamental questions on the specifics of this effect remain topics of discussion. It is not clear which forms of memory are affected by sleep and whether this beneficial effect is partly mediated by passive protection against interference. Moreover, a putative correlation between the structure of sleep and its memory-enhancing effects is still being discussed.

Design:

In three experiments, we tested whether sleep differentially affects various forms of declarative memory. We varied verbal content (verbal/nonverbal), item type (single/associate), and recall mode (recall/recognition, cued/free recall) to examine the effect of sleep on specific memory subtypes. We compared within-subject differences in memory consolidation between intervals including sleep, active wakefulness, or quiet meditation, which reduced external as well as internal interference and rehearsal.

Participants:

Forty healthy adults aged 18–30 y, and 17 healthy adults aged 24–55 y with extensive meditation experience participated in the experiments.

Results:

All types of memory were enhanced by sleep if the sample size provided sufficient statistical power. Smaller sample sizes showed an effect of sleep if a combined measure of different declarative memory scales was used. In a condition with reduced external and internal interference, performance was equal to one with high interference. Here, memory consolidation was significantly lower than in a sleep condition. We found no correlation between sleep structure and memory consolidation.

Conclusions:

Sleep does not preferentially consolidate a specific kind of declarative memory, but consistently promotes overall declarative memory formation. This effect is not mediated by reduced interference.

Citation:

Schönauer M, Pawlizki A, Köck C, Gais S. Exploring the effect of sleep and reduced interference on different forms of declarative memory. SLEEP 2014;37(12):1995-2007.

INTRODUCTION

After encoding, newly acquired memories undergo a phase of consolidation, during which the memory trace can be stabilized and strengthened. It has been shown that sleep facilitates this process.¹ Although the relation between sleep and memory has already been investigated by several generations of scientists,^2–5 fundamental questions remain unanswered.

First, it has been suggested that only specific types of declarative tasks are affected by sleep. For example, it has been proposed that semantically unrelated word lists but not semantically related ones benefit from sleep,⁶ whereas other studies showed effects using related material.⁷ Similarly, some authors suggest that benefits from sleep are greater for weaker compared to stronger memory traces,⁸ whereas other studies—using similar verbal material—report that only individuals who show a strong initial encoding benefit from subsequent sleep.⁹ Furthermore, recent studies have found that future relevance¹⁰ might mediate the effect sleep has on memory consolidation, but that future reward does not.¹¹ Sometimes, sleep related changes can only be found in brain activity, but not in overt changes in behavioral performance.¹² Finally, it has been suggested that mainly the types of memory that rely on the hippocampus are sleep dependent.^5,13,14 Therefore, it has been proposed that the mode of retrieval could also mediate whether memory performance is influenced by sleep, as recollection-based memory is thought to depend more on the hippocampus than familiarity-based memory.^15,16 Currently, it is unclear whether any of these types of memory tasks is actually not sleep dependent.

Second, it is not yet clear whether the positive effect of sleep on declarative memory formation stems from an active role of sleep in consolidation or whether it is passively emerging from reduced interference during an interval spent asleep as compared to one spent awake. In the procedural domain, this matter has already been addressed. To reduce interference generated by motor activation, Walker et al.¹⁷ stabilized the participants' arms during consolidation of a motor skill learning task, so that there were no differences in muscle activity between wake and sleep groups. Similarly, Mednick et al.¹⁸ tested whether the benefit of sleep for a visual texture discrimination task resulted from absence of visual input in the sleep group by adding a control group that was blindfolded while awake. Both studies show that sleep, but not interference-free wakefulness, improved procedural memory consolidation.

Establishing an interference-free control for a declarative memory task is not as straightforward as blindfolding subjects and letting them rest comfortably. Deliberate rehearsal can occur even in the absence of any external stimulation and leads to improved memory performance. Random cogitation, however, can constitute interference. An interference-free wake control for a declarative task must therefore involve not only suppression of all external sensory stimuli but also of internal stimuli. To control external stimulation, patients can be seated in a dark and quiet room while their state of consciousness is monitored by electroencephalography (EEG). An efficient way to control cognition is by focusing attention on one single thought, limiting self-generated information processing to a minimum. This type of concentration is strongly developed in people who practice meditation. Meditators have learned to focus their thoughts for a very long time,¹⁹ thus reducing cognitive interference from internal stimuli most effectively and constituting a suitable low-interference control. A third topic of ongoing discussion is the question whether the strength of sleep dependent memory consolidation is mediated by the amount of time spent in a specific sleep stage, i.e., by the “macrostructure” of sleep.^9,20–22 For example, declarative memory has been proposed to benefit mainly from slow wave sleep (SWS).¹ However, the amount of time spent in sleep stage 2 has been found to predict declarative memory consolidation.²³ Other studies focus on sleep “microstructure,” i.e., sleep related mechanisms that could mediate the effect on declarative memory consolidation. Sleep spindles, brief bursts of synchronous neuronal firing in the frequency range of 12–16 Hz, have been shown to favor synaptic plasticity.²⁴ It has also been shown that previous learning experience modulates spindle activity during sleep²⁵ and that sleep spindle activity correlates with the strengths of the observed benefits on memory consolidation or memory performance.^{10,23,26–28} Recently, sleep spindles have also been found to be a pacemaker for memory reactivation.²⁹ Furthermore, several studies show the functional importance of slow wave activity for sleep dependent declarative memory consolidation.^30,31 Even though several models regarding effects of different sleep stages on memory consolidation have been proposed,^7,32 we agree with a recent paper concluding that “no model for the sleep-stage dependencies of memory processing has been found to adequately explain the existing data”, and that results of correlational studies currently remain inconclusive.³³

We tried to systematically address these open questions regarding the nature of the declarative sleep effect in a series of experiments. We tested whether differences in learning material can explain previous inconsistent findings. First, we tested a broad range of declarative memory tasks, varying in type of material and recall mode. Second, we investigated whether sleep holds an active role in declarative memory consolidation by comparing an active wake condition, a reduced-interference wake condition, and a sleep condition. Third, we studied the influence of sleep macrostructure and microstructure by comparing sleep with varying amounts of time spent in different sleep stages and analyzing the relationships between sleep parameters and memory consolidation.

MATERIALS AND METHODS

Test Material

A standard diagnostic test battery was used to test effects of sleep on different types of declarative memory (“Lern- und Gedächtnistest”, LGT-3³⁴). The LGT-3 is a speeded memory test commonly applied in German performance diagnostics. The LGT-3 measures long-term memory performance over a broad range of declarative material and is constructed as a parallel test with two equally difficult versions (A and B), allowing within-subject testing of performance after the retention interval in sleep as compared with wakefulness. For Experiment 3, which tested three conditions within each subject, a third parallel version, C, was created. Versions were counterbalanced between experimental conditions. Equal level of difficulty was confirmed in a pretest sample of 10 subjects. Each parallel version consists of six subtests: “city map,” “vocabulary,” “objects,” “phone numbers,” “story details,” and “signs.” These subtests are always presented in a fixed, standardized order during learning and recall. The city map: remembering a path between two points on a map of a complex maze within 1 min. For retrieval, the path has to be redrawn on the map from memory within 2 min. Vocabulary: learning 20 Turkish-German word pairs within 1 min. For retrieval, the German words, along with five Turkish words to choose from, are presented within 4 min. Remembering objects: remembering 20 simple objects within 1 min, free recall in 2 min. Phone numbers: association of 13 three-digit numbers with names within 2 min, recall of the numbers cued with the names within 2 min. Details of a story: reading a story within 1 min, answer 14 questions concerning numbers and names in the text within 4 min. Signs: learning abstract and concrete drawings inside different frames within 1 min. For retrieval, the drawings and four frames to choose from are given within 4 min. We followed all procedures as described in the manual of the LGT-3, except for the length of the retention interval, which was adapted to fit our sleep and wake intervals. Performance in the LGT-3 is measured as number of correct items in all sub-tests, as defined in the test manual. Raw scores for all subtests is transformed to T-scores (mean: 50, standard deviation [SD]: 10) according to standard norms.

The subtests were subsumed into mutually exclusive scales, classified as verbal (vocabulary, story details) and nonverbal (city map, signs), single-item learning (objects) and item- association learning (vocabulary, phone numbers, signs), recall (city map, objects, phone numbers, story details), and recognition (vocabulary, signs), as well as cued recall (phone numbers, city map) and free recall (objects, story details). Additionally, a total score is computed according to the LGT-3 test manual,³⁴ consisting of the sum of T-transformed scores of all six subtests. This score serves as a general measure of declarative memory performance. Classification into mutually exclusive scales followed description of demands on memory operations given in the test manual. Only subtests that belonged unambiguously to one category were included in a scale.

General Design and Procedure

Sixty-one paid volunteers recruited from the Ludwig-Maximilians-Universität and different Buddhism centers in Munich participated in three experiments. They were healthy, 18–55 y old, nonsmokers, and had no Turkish skills. They did not ingest any medication, alcohol, or caffeine on the days of the experiments. All participants reported sleeping between 6 and 10 h per night, had a regular circadian rhythm, and were not extreme morning or evening types, as confirmed by the Munich Chrono-type Questionnaire (MCTQ).³⁵ They had no shift work or long-distance flights within the 6 w preceding the experiment, and experienced no sleep related pathology.

In all experiments, subjects learned a battery of declarative memory tasks (LGT-3)³⁴ before a period of sleep or wakefulness. After this period, the learned material was tested. No immediate test was given after learning, because the LGT-3 does not permit more than one testing per parallel test version. Experiment 1 compared afternoon periods of sleep (with a higher proportion of SWS) and wakefulness. Experiment 2 compared morning periods of sleep (with a higher proportion of rapid eye movement [REM] sleep) and wakefulness. In Experiment 3, highly trained meditators participated in a sleep condition, an active wake condition during which they were involved in discussion with the experimenter, and an interference-reduced quiet meditation condition. Sleep opportunity in all experiments was 2 h. Subjects showing less than 20 min of sleep or not reaching S2 were excluded from analysis. All experiments used within-subject comparisons. To facilitate daytime sleep, participants were asked to rise 1 h earlier than usual the two mornings before all conditions. Sleep logs were used to confirm that participants kept to the administered sleep schedule and to control the sleep history. Subjects had the same amount of sleep in the nights preceding the different experimental conditions (sleep 6 h 23 min ± 1 h 7 min [mean ± SD], wake 6 h 34 min ± 1 h 1 min, meditation 6 h 47 min ± 1 h 2 min; all P > 0.31 for pairwise comparisons within experiments and in combined data), and there were no differences in average sleep times during the week preceding the experimental sessions (sleep 7 h 07 min ± 54 min, wake 7 h 11 min ± 46 min, meditation 6 h 54 min ± 1 h 11 min, all P > 0.22 for pairwise comparisons within experiments and in combined data). Our experimental design used short daytime sleep windows as an additional sleep opportunity to exclude confounding effects of sleep loss on memory recall performance. Daytime naps have been found to have no quantitative effect on levels of fatigue and alertness, as measured by the psychomotor vigilance task (PVT).^36,37 This makes daytime sleep an ideal condition to compare consolidation effects on memory. In our experiments, we controlled alertness using the standard 10-min version of the PVT.³⁶ The PVT was administered after the memory test session, following sleep, wakefulness, and meditation. There were no significant differences between conditions in the most sensitive measure number of lapses (reaction times > 500 msec)³⁸ nor in median response times (all P > 0.33). The minimum time between two experimental conditions was 1 w in order to avoid interference between sessions. The order of conditions and the order of parallel LGT-3 test versions were counterbalanced in all three experiments. In addition, we confirmed that overall test performance did not differ between the first and second test session in all three groups (all P > 0.35 for total score differences; see Table S1 in the supplemental material for more detailed information regarding all scales).

Experiment 1. A sample of 20 students (10 male), aged between 18 and 30 y (21 ± 3 y [mean ± SD]), participated in the experiment. One participant was excluded because she did not fall asleep during the sleep condition. For one participant, not all memory scores could be calculated because of a missing subscale. EEG data were missing for two participants because of technical problems during the measurements. Participants arrived at the sleep lab at 13:00. After learning the LGT-3 memory test battery (15 min) and application of EEG electrodes (30 min), subjects either had the opportunity to sleep for 2 h or had to stay awake until testing. At 16:15, memory retention was tested (30 min; see Figure 1A). In the wake condition, participants played easy, nonverbal board games with the experimenter.

Figure 1.

Experimental design. We compared the effect of sleep on memory consolidation in three different experiments. The first (A) and second (B) experiment tested sleep at different times of day and with different delay between learning and testing. Subjects slept either in the afternoon (A,C) or in the morning (B). In the first and third (C) experiment, subjects were tested after 3 h, whereas there was an 8-h retention interval in the second experiment (B). The third experiment (C) compared an additional interference-reduced wake control with sleep and active wakefulness, where participants had to meditate for 2 h, directing their focus of attention on a single object, such as their breath or a mantra. L, learning; R, recall.

Experiment 2. A sample of 20 students (7 male), aged between 18 and 30 y (24 ± 4 y), participated in the experiment. Two participants were excluded because they did not fall asleep during the sleep condition. Participants arrived at the sleep laboratory at 08:00. After learning the LGT-3 memory test battery (15 min) and application of EEG electrodes (30 min), subjects either had the opportunity to sleep for 2 h or had to stay awake until testing. Subjects slept in the laboratory, but were allowed to leave for the remaining time until memory retention was tested at 16:15 (see Figure 1B). During the time spent outside the sleep laboratory, activity was monitored by actimetry.

Experiment 3. Seventeen highly trained meditators (14 male), aged between 24 and 55 y (40 ± 10 y), with an average meditation experience between 2 and 28 y (10 ± 8 y), participated in the experiment. Participants were required to have at least 2 y of meditation experience with daily meditation practice to be included in the study. Two subjects were excluded because they slept less than 20 min in the sleep condition, and one subject was excluded, because he fell asleep during meditation. In one subject not all memory scores could be calculated because of a missing subscale, and one subject had to be excluded because of exceptionally low learning performance in all conditions, leaving a total of 12 subjects.

Participants arrived at the sleep laboratory in the afternoon between 12:00 and 17:00, at the same time for all three study conditions. After learning the LGT-3 memory test battery (15 min) and application of EEG electrodes (30 min), subjects were assigned to one of three conditions. To keep procedures in all three conditions identical, EEG electrodes were applied also in the active wake condition, where they were not required. Afterward, in the sleep condition, subjects had the opportunity to sleep for 2 h. In the active wake condition, they discussed various topics with the experimenter while sitting in a busy street café. In the interference-reduced wake condition, they meditated for 2 h in a quiet room. During meditation, participants directed their focus of attention on a single object such as their breath or a mantra. Four hours after learning, memory retention was tested (see Figure 1C). Meditation was monitored with EEG.

Statistical Analysis

Because retrieval performance differed to some extent between the three experiments because of different length of the retention interval and different age groups, all LGT-3 scales were standardized to z-scores (mean: 0, SD: 1) separately for the experimental groups. Thus, effects of conditions and scales are made comparable between experiments. In addition, this transformation allows results of all three experiments to be pooled and analyzed as one large sample. Note that all experiments used within-subject comparisons. Thus, differences between experimental conditions (sleep, wake, meditation) are preserved by this transformation. T-tests for dependent measures and repeated-measures analyses of variance (ANOVAs) with a two-sided significance threshold of 0.05 were used to compare performance between conditions. Results are given as mean ± standard error of the mean (SEM). Correction for multiple testing was done using the Holm-Bonferroni method, which adjusts α according to a stepwise procedure that no longer corrects for the tests that survived a stricter Bonferroni correction in previous steps. The Holm-Bonferroni method adjusts the corrected significance threshold α to n − (k − 1), where n is the total number of tests performed and k is the rank of the test among all tests ordered from lowest to highest P value. Tests are considered significant according to the chosen threshold α as long as P < α / (n − (k − 1)).

EEG Recording and Analysis

In all experiments, sleep was recorded polysomnographically in the sleep laboratory. EEG (C3 and C4, according to the international 10-20 system) against a nose reference, and bipolar electrooculogram and electromyogram were recorded. Data were recorded and scored offline by two independent raters according to the Rechtschaffen and Kales standard criteria.³⁹ Discrepancies in scorings were decided by a third rater. Meditation was monitored with the same setup to ascertain that no sleep occurred in this condition.

For spectral analysis, EEG data was segmented into 1-sec epochs and artifacts were rejected using a semiautomatic process that detects muscle activity, signal jumps, and bad channels. Only artifact-free data were entered into analysis. Log-transformed band power was calculated in the frequency range of slow oscillations (0–1 Hz), delta waves (1–4 Hz), and sleep spindles (12–16 Hz). Sleep spindles were determined and counted by a previously published algorithm.²⁵ The individual peak frequency was detected in the spindle frequency range (12–16 Hz) and EEG data were bandpass-filtered in the individual spindle range for each subject (peak frequency ± 1.5 Hz). Spindles were defined as a root-mean-square signal that stayed above an individual, SD-based threshold (1.5 SDs above average power) for 0.5 to 3 sec, with at least 0.5 sec between events. Both number of spindles and spindle activity [duration × amplitude] were calculated.

RESULTS

Behavioral Data

In all three experiments, we tested whether sleep benefits the consolidation of different kinds of declarative memory equally. In Experiment 1, subjects learned in the afternoon and were tested after a 3-h retention interval during which they took a 2-h nap in the sleep condition. They performed significantly better after sleep than after wakefulness on a number of scales (for inferential statistics, see Table 1). The benefit from sleep as measured by sleep-wake differences was significant for non-verbal memory, for memory for associated items, and when recognition was tested. Sleep did not yield a significant benefit for verbal memory, for memory for single items, or when recall was tested, regardless of whether cued recall or free recall was tested. In addition, overall performance as measured by the sum score of the test battery was significantly better in the sleep than in the wake condition (see Figure 2A). However, it is important to note that our data do not show that sleep preferentially affects specific types of declarative memory. Interaction analyses tested whether in pairs of mutually exclusive task dimensions (verbal versus nonverbal [ve−nv], single item versus associated items [si−as], recall versus recognition [rc−rg], cued recall versus free recall [cr−fr]) one type of memory benefitted more from sleep (s) than the other type. All interactions remained nonsignificant (ve−nv × s−w: F_1,17 = 0.73, P = 0.41; si−as × s−w: F_1,17 = 0.86, P = 0.37; rc−rg × s−w: F_1,17 = 1.24, P = 0.28; cr−fr × s−w: F_1,17 = 0.42, P = 0.53; Table 1). In the aforementioned analyses, the subtests of the LGT-3 were subsumed into scales according to type of memory. When looking at single subtests, subjects performed significantly better in the sleep than in the wake (w) condition in the signs test (see Table 1).

Table 1.

Mean z-differences (sleep - wake) for scales and single subtests.

Figure 2.

Differences in memory scores between the sleep and wake condition for the verbal (ve), the nonverbal (nv), the single item (si), the associate (as), for recall (rc), recognition (rg), cued recall (cr), and free recall scale (fr) for the sleep and wake condition for A, the first experiment (n = 18), B, the second experiment (n = 18), C, the third experiment (n = 12) separately, and D, for all data combined (n = 48). * P = 0.05, ** P = 0.01, *** P = 0.001.

In Experiment 2, subjects learned in the morning, then took a 2-h morning nap in the sleep condition and stayed awake until they were tested 8 h after encoding. They again performed significantly better after sleep than after wakefulness on a number of scales (see Table 1). Significant sleep-wake differences were found for verbal memory, for memory for single items, and when recall was tested, in particular when a free recall procedure was used. Sleep did not yield a significant benefit for nonverbal memory, for memory for associated items, when recognition was tested, or when cued recall was tested. In addition, overall performance as measured by the sum score of the test battery was significantly better in the sleep condition (see Figure 2B). Again, it is important to note that we find no sign that sleep affects one form of memory more than another. For all mutually exclusive dimensions, interactions with state (sleep/wake) remained nonsignificant (ve−nv × s−w: F_1,17 = 2.00, P = 0.17; si−as × s−w: F_1,17 = 2.22, P = 0.16; rc−rg × s−w: F_1,17 = 2.51, P = 0.13; cr−fr × s−w: F_1,17 = 0.58, P = 0.46; Table 1). Again, in the aforementioned analyses, the subtests of the LGT-3 were subsumed into scales. Looking at single subtests, subjects performed significantly better in the sleep than in the wake condition in the vocabulary and objects tests (see Table 1).

In Experiment 3, three conditions (sleep/active wake/meditation) were compared. Procedures corresponded to those of Experiment 1. Subjects learned in the afternoon and slept or meditated for 2 h in the respective conditions during the 3-h retention interval. Conditions differed on a number of scales; and post hoc tests confirmed that subjects were significantly better in the sleep condition than in the two wake conditions (see Table 2). We found a significant main effect of condition for verbal memory, for memory for single items, and when free recall was tested. Post hoc t-tests confirmed that subjects performed significantly better in the sleep condition than in the wake or meditation (m) conditions for verbal memory (s−w: t₁₁ = 3.17, P = 0.01, s−m: t₁₁ = 2.79, P = 0.02), for memory for single items (s−w: t₁₁ = 2.48, P = 0.03, s−m: t₁₁ = 2.35, P = 0.04), and when free recall was tested (s−w: t₁₁ = 2.93, P = 0.01, s−m: t₁₁ = 2.89, P = 0.02). There were no significant main effects for nonverbal memory, for memory for associated items, when recognition was tested, when recall in general was tested, or when a cued recall procedure was used. In addition, overall performance as measured by the sum score of the test battery showed a significant main effect. Post hoc t-tests confirmed that subjects performed significantly better in the sleep condition than in the wake or meditation conditions (s−w: t₁₁ = 2.34, P = 0.04, s−m: t₁₁ = 3.45, P = 0.01; see Figure 2C and Figure 3).

Table 2.

Mean z-values for scales and single subtests in experiment 3.

Figure 3.

Memory scores for the sleep and wake and meditation condition in the third experiment for the verbal (ve), the nonverbal (nv), the single item (si), the associate (as), for recall (rc), recognition (rg), cued recall (cr), and free recall scale (fr). N = 12. Asterisks mark significant difference to the sleep condition. * P = 0.05, ** P = 0.01.

As in Experiments 1 and 2, interactions between state (sleep/wake) and test dimension showed no sign that sleep differentially affects different forms of memory (ve−nv × s−w−m: F_2,22 = 1.05, P = 0.37; si−as × s−w−m: F_2,22 = 0.80, P = 0.46; rc−rg × s−w−m: F_2,22 = 0.05, P = 0.95; cr−fr × s− w−m: F_2,22 = 1.76, P = 0.20). To test whether interference-reduced wakefulness differed from active wakefulness, post hoc tests were calculated also between the meditation and wake conditions. These remained nonsignificant for all dimensions (ve: t₁₁ = 0.70, P = 0.50, nv: t₁₁ = 0.01, P = 0.99, si: t₁₁ = 0.37, P = 0.72, as: t₁₁ = 0.08, P = 0.94, rc: t₁₁ = 0.31, P = 0.76, rg: t₁₁ = 0.05, P = 0.96, cr: t₁₁ = 0.53, P = 0.61, fr: t₁₁ = 0.02, P = 0.99; Table 2). There was also no difference in performance on the total score between the wake and meditation condition (t₁₁ = 0.13, P = 0.90).

Again, in the aforementioned analyses, the subtests of the LGT-3 were subsumed into scales. Looking at single subtests, subjects performed significantly better in the sleep than in the wake condition in the vocabulary and objects tests (Table 2).

Finally, we pooled data from all 48 subjects in a combined analysis of all three experiments. A multivariate ANOVA over all scales revealed no interaction between experiment and condition (F16,78 = 0.77, P = 0.71, all univariate comparisons P > 0.17), whereas there was a highly significant main effect of condition (sleep versus wake) over all scales (F_8,38 = 3.7, P = 0.003). This confirms that differences in experimental procedures (time of day, subjects' age, etc.) did not significantly influence the effect of sleep. Furthermore, individual ANOVAs confirmed that sleep benefited memory performance significantly for all but one sub-scale (Table 1). Significant sleep-wake differences were found for verbal memory, for memory for single items, for memory for associated items, when memory recall was tested, when either a cued recall procedure, or a free recall procedure was used, and also when memory recognition was tested. Sleep did not yield a significant benefit for nonverbal memory tests, although a numerically higher value was found in the sleep condition. Finally, overall performance as measured by the sum score of the test battery was significantly better in the sleep condition (Figure 2D). Again, also in this larger sample with higher statistical power, we found no indication that sleep affects some forms of memory more than others. There was no significant interaction between state (sleep/wake) and any of the mutually exclusive dimensions tested (ve−nv × s−w: F_1,47 = 1.17, P = 0.29; si−as × s−w: F_1,47 = 0.38, P = 0.54; rc−rg × s−w: F_1,47 < 0.01, P = 96; cr−fr × s−w: F_1,47 < 0.01, P = 0.94; Table 1). Effect sizes (Cohen d) for the scales range between 0.3 and 0.5, translating to small or medium sized effects (ve: d = 0.47, nv: d = 0.26, si: d = 0.44, as: d = 0.46, rc: d = 0.52, rg: d = 0.42, cr: d = 0.34, fr: d = 0.48, total score: d = 0.51).

Looking at single subtests, without subsuming them into scales, subjects performed significantly better in the sleep than in the wake condition in the tests vocabulary, objects, phone numbers, and story details. Sleep did not yield a significant benefit for the city map and signs tests (Table 1).

Although significant differences do not emerge for the same memory scales and individual subtests in the three experiments, the overall sum test score was significant for all three experiments. To further demonstrate the stability of this effect, Figure 4 shows that, on an individual level, most subjects show equal or better performance in the sleep than the wake conditions.

Figure 4.

Performance of individual subjects on the LGT-3 total score in all three experiments. Note that this figure shows raw scores instead of z-transformed scores for individual subjects. Almost all subjects show better performance in the sleep compared with the wake condition (A, B) or the average of active wake and meditation conditions (C). A, In Experiment 1, only one of 18 subjects was better after an interval spent waking. B, In the second experiment two of 18 subjects performed slightly worse in the sleep condition. C, In Experiment 3, again only one subject in 12 does not show a beneficial effect of sleep. Subjects performed better on the test when there were only 3 h between learning and testing (A) than when encoding and testing were spaced 8 h apart (B). The older subject pool (C), on average, performs not as well as the younger subjects who had the same delay between learning and testing (A). Still, we observe a significant benefit of sleep in all three populations, indicating that initial performance does not have an effect on whether sleep boosts memory consolidation. Dashed lines, subjects who did not show slow wave sleep (SWS); dotted lines, subjects who did not have rapid eye movement (REM) sleep during the nap, dash-dotted lines, subjects who neither had SWS nor REM sleep; solid lines, subjects who had both SWS and REM sleep during the sleep interval.

EEG Data

As intended in the experimental design, the amount of time spent in different sleep stages differed significantly between Experiments 1, 2, and 3. Whereas afternoon sleep in Experiment 1 had a higher proportion of SWS, morning sleep in Experiment 2 showed more REM sleep. Subjects in Experiment 3 had generally less deep sleep and fewer sleep spindles, as can be expected from the older subject population (Table 3). Because the difference between sleep and wakefulness was most robust for the sum scores of the test battery, we correlated this measure with the amount of time spent in different sleep stages. No significant correlations between sleep macrostructure (time spent in sleep stages 1, 2, SWS and REM) and the sum score of the test battery were found (Table 4). The amount of time spent in specific sleep stages did not correlate with any measure of memory performance (for 144 correlations of eight sleep parameters and nine scales, both with the difference between sleep and wake and with performance in the sleep condition alone, all P > 0.08). In addition, we also investigated whether subjects who did not have SWS or REM sleep still showed sleep related improvements. The individual traces in Figure 4 show that the effect of sleep is not limited to those subjects who show all sleep stages. In fact, chi-square tests confirm that the distribution of subjects showing sleep-related improvement is equal among subjects with and without SWS/ REM sleep (SWS: χ² = 0.55, P = 0.46; REM sleep: χ² = 0.70, P = 0.40).

Table 3.

Sleep parameters in the different experiments.

Table 4.

Correlations between sleep parameters and performance on the LGT total score.

To investigate the influence of sleep microstructure on memory consolidation, we correlated spindle activity and EEG band power in the slow oscillation and delta wave range with the overall sum score of our test battery. We found a correlation between spindle activity in stage 2 and the total score of the LGT-3. This correlation, however, was higher and only reached significance with performance in the wake condition (sleep: r₄₆ = 0.11, P = 0.46, wake: r₄₆ = 0.38, P = 0.009; see Figure 5 and Table 4). Sleep related improvement (difference sleep-wake), however, showed a marginally significant negative correlation with spindle activity (r₄₆ = -0.28, P = 0.06). Thus, higher spindle power during sleep seems to be associated with general performance in the LGT-3 (trait), but not the memory benefit related to sleep. Band power in the range of slow oscillations and in the delta frequency range did not predict performance after sleep, wakefulness, or the benefit achieved by consolidation in sleep (all P > 0.71, see Table 4).

Figure 5.

Correlation of electroencephalographic sleep spindle activity with LGT-3 performance. A, Spindle activity measured in the sleep condition shows a significant positive correlation with performance on the LGT-3 in the wake condition (r₄₆ = 0.38, P = 0.009). B, Sleep spindle activity shows a weaker association with performance after sleep, with this correlation remaining non-significant (r₄₆ = 0.11, P = 0.46). Thus, present data indicate that spindle activity is rather a trait than a state marker of memory performance. NREM, nonrapid eye movement sleep.

DISCUSSION

In three different experiments, we tried to address some of the unanswered questions regarding the effect of sleep on declarative memory. Although the first experiments on this topic were performed 100 y ago,² it is still not known whether the effect seen occurs for all types of declarative memory tasks. It has also been debated for a long time whether sleep actively contributes to consolidation or whether it simply protects memory from interference. Finally, it has been proposed that specific sleep stages (“macrostructure”) or features (“microstructure”) are of importance for memory consolidation; yet, no definite relations have emerged thus far. We used a comprehensive test battery to test whether sleep has a differential effect on various types of declarative memory. The test battery contained scales with verbal and nonverbal material, associative and nonassociative items, and different types of recall. We found that sleep consistently promotes declarative memory consolidation over the entire range of tasks, if an adequate sample size is used. A combined measure of different declarative memory scales showed consistent sleep effects also in smaller samples. This enhancing effect is restricted to sleep, and does not occur in quiet wakefulness. There is no difference between an active wakefulness condition, in which subjects were in a busy environment and actively discussed different topics with an experimenter, and a condition in which subjects reduced external and internal interference by focusing their attention on a single thought in meditation. Thus, it is reasonable to assume that the underlying process is sleep-specific, active in nature, and not passively mediated by reduced interference. This beneficial effect of sleep seems not to be related to the time spent in a certain sleep stage, nor by sleep microstructure.

Different Types of Memory and Sleep

We find a positive effect of sleep on declarative memory for a variety of different memory tasks, independent of the material used and by which method it was retrieved.

Although not all scales were significant in all three samples, pooling all 48 subjects reveals significant effects in nearly all tests. We conclude from this observation that the extent and variability of the effect of sleep is not large enough to be consistently detected in sample sizes of fewer than 20 subjects, which are habitually used in sleep research. In particular, experiments showing conditions with no effect must be interpreted with great caution, especially if no significant sleep × condition interaction is present. Our results show that the standardized effect size is between 0.3 and 0.5, which translates to small to medium effects according to Cohen.⁴⁰ To detect an effect of this size with a statistical power of at least 0.5, (i.e. with the ability to detect a significant effect at least in every second experiment,) a sample size between 20 and 50 per condition is required. This estimate is considerably smaller than that of other studies, which rely on smaller sample sizes. It might be partly accounted for by our experimental design. For example, we use naps, which probably produce smaller effects than full nights of sleep, although some nap studies also report larger effect sizes.⁴¹ However, a systematic investigation of the size of the sleep effect on memory has not been done, and any estimate of effect size, which does not consider the publication bias against null results, will result in overestimation of effects. If we want to systematically investigate the relation between sleep and memory, and not only focus on the odd and rare findings with exceptionally large effects, sample sizes have to increase considerably.

Our results also suggest that it can be of advantage to combine a number of different memory scales when trying to measure sleep related consolidation. No subscale was significant in all three experiments. However, the total test score always reached significance. It should therefore be considered to use a broader spectrum of tests particularly when the effects of sleep in more realistic settings are to be studied and a precise differentiation between memory tasks is of lesser importance. Effects of interference among multiple tests during encoding and their consequences for subsequent memory consolidation, however, are not well characterized and should be addressed in future studies.

We tested whether material specific differences in sleep related memory consolidation exist, and found none. This means that the processes in sleep occur fairly generally for declarative memory. There might still be differences between specific types of memory, even though we did not find them in our experiments. We find similar effects on verbal as well as on nonverbal memory, and on single-item memory as well as on associated items. In the literature, effects have been reported on emotional as well as on neutral material.^42–46 Wordlist learning of both related^7,10,47–49 and unrelated word pairs^6,9,50,51 has been shown to benefit from sleep. Yet, there are some indications that only specific material benefits from sleep.^6,23 When combining data across experiments, all tested scales reached significance except for nonverbal memory. The interaction effect between verbal and nonverbal memory, however, was not statistically significant. In the literature, reports of significant effects of sleep on nonverbal memory are rare, maybe because it is less frequently tested, or maybe because effect sizes are smaller. Many authors find effects of sleep only on consolidation of emotional pictures,^43,52 but only few find the same effect for emotionally neutral pictures.⁴⁴ In view of the aforementioned considerations on statistical power, and taking into account the compiled literature, it is probably too early for final conclusions on how specific types of declarative memory are affected by sleep.

The mode of memory retrieval was, in the current experiments, not relevant for whether sleep had an effect on memory performance. This is in line with a number of studies showing sleep effects with free recall,^41,53 cued recall,^9,51,54 or recognition.^20,55,56 However, there are several studies indicating that only the recollection and not the familiarity aspect of recognition memory is enhanced by sleep.^15,45,57 This would speak for a particular influence of sleep on hippocampal memory because only recollection, but not familiarity, is supposed to depend on the hippocampus proper.⁵⁸ Our finding is not in conflict with these studies because our test battery was not designed to separate familiarity and recollection.

Because the LGT-3 tests various types of declarative memory, we cannot exclude interference between tasks. However, although subtests were always presented in the same order during learning, we did not observe preferential consolidation of any of the studied subtests. This is consistent with the finding that sleep after learning removes effects of retroactive interference.⁸ Additionally, because interference between tasks was identical over conditions, our reported findings cannot be attributed to effects of task interference.

An Active or Passive Role of Sleep

Interference can disturb previously learned memories. Because of thalamic gating, interference during sleep is reduced to a minimum. Thus, the question arises whether sleep produces its beneficial effect passively by preventing interference, or actively by strengthening new memory traces.^59,60 Early work by Ekstrand and colleagues⁴ examined the effect of interference on memory, but did not come to a final decision whether interference or other processes mediate the sleep effect. Currently, active and passive views of sleep are discussed,^1,61,62 although evidence from many directions points to an active role of sleep. Animal and human studies show reactivation of neural activity patterns in the hippocampus and neocortex during sleep that can be related to previous learning.^63–66 External reactivation by presenting memory cues during sleep can selectively enhance the consolidation of individual memories⁶⁷ and also accelerates the consolidation process.⁶⁸ Other studies tested whether interference before or after the consolidation interval interacted with the effect of sleep. Ellenbogen et al.⁵⁰ tested whether interference leads to similar memory deterioration after a period of sleep and after a period of wakefulness. Their results show that sleep makes the memory trace less susceptible to interference. This can be taken as evidence of an active consolidation process during sleep. Drosopoulos et al.⁸ showed that sleep also rescues memory from retroactive interference that occurred directly after learning. This, again, speaks for an active role of sleep.

The current study concerned itself with the question whether reduced interference during the retention interval contributes to the beneficial effects of sleep on declarative memory. Reduced interference during wakefulness did not improve memory consolidation. Potentially, interference may still have occurred during the 30-min electrode application period following learning. However, this experimental procedure was the same for all the conditions, including sleep and active wakefulness. Hence, any interference during this period cannot account for our findings.

Reducing interference during the retention period has been shown not to interact with the effect of sleep on procedural memory.^17,18 The current experiments did not find a difference between an active and a quiet wake condition for declarative memory either. This is also remarkable because subjects, being expert meditators, were admittedly motivated to show maximal performance in the meditation condition. If reduced interference, increased motivation, or potential rehearsal had any influence on performance, the meditation condition should show improvement over the active wake condition, which was not the case. Therefore, in addition to providing an interference-free period, sleep must have additional properties that support memory consolidation. These could be related to the switching between neuromodulatory states, which happens when the organism falls asleep, or they could be linked to electrophysiological characteristics of sleep, such as sleep spindles and delta waves. Although we could not find any correlations, such effects have been reported in other studies.^{33,56,69–71} The exclusivity of some consolidation processes to sleep is also supported by the fact that external reactivation only benefits memory consolidation when subjects sleep, not when they were awake.^72,73 Sleep thus seems to hold a special and active role in the consolidation of declarative memory, which may not be explained by lack of interference during the consolidation interval.

Macrostructure Versus Microstructure of Sleep

Another point of debate is whether specific sleep stages (sleep macrostructure) contribute particularly to memory consolidation, or whether specifics in sleep physiology (sleep microstructure) are at the base of consolidation.⁷⁴ It has often been suggested that the amount of time spent in a particular sleep stage mediates the positive effect of sleep on memory formation.⁷⁵ However, no consensus on this point has been reached.³³ In the current study, although drawing on a large sample with 46 subjects and sufficient variance in sleep and in memory performance, no correlation between sleep structure and sleep related memory improvement has been found. Apart from the missing correlation, another argument against an influence of sleep macrostructure on memory consolidation is that the amount of time spent in different sleep stages in our three experiments varied because morning and afternoon naps were investigated, which showed higher amounts of REM sleep and SWS, respectively. However, no consistent pattern emerged that could relate time spent in different sleep stages with specific types of memory, and even subjects without SWS or REM sleep show sleep related improvements.

Sleep spindles and slow waves, which are the neural mechanisms associated most often with memory consolidation, also show no correlation with sleep related memory benefits in our data. Literature suggests that sleep spindles can be markers of trait-like and state-like performance indices.⁷⁴ We therefore correlated spindle activity with performance in the wake condition (trait) and with performance increase in the sleep condition (state). The only significant relationship between sleep microstructure and performance is a positive correlation between spindle activity and memory performance in the wake condition. Thus, spindle activity during sleep seems to be a marker of a trait-like feature. The correlation we report does not survive a very conservative Bonferroni correction, which accounts for all tests performed on all correlations, and should thus be treated with caution. Yet, it remains significant when using the more common correction only for the independent comparisons of each group of tests and is the strongest association of sleep parameters and performance in this large dataset. Our finding is in line with a number of studies that show an association between spindle activity and general cognitive abilities.^76–79 It must, however, be noted that our experiments do not exclude causal relationships between sleep microstructure and memory processing that extend beyond simple quantitative association.²⁵ In fact, we believe that consolidation processes are rather related to changes in neurotransmitter activity and memory trace-specific reactivation, both of which are probably not directly reflected in global measures of sleep.

CONCLUSIONS

First, we show that sleep consistently benefits the consolidation of declarative memory. We do not find that only certain types of declarative memory benefit from sleep, but that a broad range of tests are promoted by sleep, if tested in a sufficiently large sample. In addition, a broader coverage of different memory tasks can also reduce the risk of type II errors and help detect the small to medium-sized effects of sleep that are usually reported. Second, consolidation was not improved in a reduced-interference wake control group. Thus, the effect of sleep on memory consolidation is active in nature and not merely caused by a lack of interference. Finally, the beneficial effect of sleep we found in all three experiments was not mediated by time spent in certain sleep stages nor by the amount of activity in a specific EEG band during sleep. Therefore, we believe that it is the intricate interplay of sleep related physiological processes that allows memory consolidation to be particularly strong during sleep.

DISCLOSURE STATEMENT

This was not an industry supported study. This study was supported by the Deutsche Forschungsgemeinschaft (DFG) grant GA730/3-1. The authors have indicated no financial conflicts of interest.

SUPPLEMENTAL MATERIAL

Table S1

Effects of session order for all scales and total score.