Sleep has been shown to promote memory consolidation driven by certain oscillatory patterns, such as sleep spindles. However, sleep does not consolidate all newly encoded information uniformly but rather “selects” certain memories for consolidation. It is assumed that such selection depends on salience tags attached to the new memories before sleep. However, little is known about the underlying neuronal processes reflecting presleep memory tagging. The current study sought to address the question of whether event-related changes in spectral theta power (theta ERSP) during presleep memory formation could reflect memory tagging that influences subsequent consolidation during sleep. Twenty-four participants memorized 160 word pairs before sleep; in a separate laboratory visit, they performed a nonlearning control task. Memory performance was tested twice, directly before and after 8 hr of sleep. Results indicate that participants who improved their memory performance overnight displayed stronger theta ERSP during the memory task in comparison with the control task. They also displayed stronger memory task-related increases in fast sleep spindle activity. Furthermore, presleep theta activity was directly linked to fast sleep spindle activity, indicating that processes during memory formation might indeed reflect memory tagging that influences subsequent consolidation during sleep. Interestingly, our results further indicate that the suggested relation between sleep spindles and overnight performance change is not as direct as once believed. Rather, it appears to be mediated by processes beginning during presleep memory formation. We conclude that theta ERSP during presleep memory formation reflects cortico-hippocampal interactions that lead to a better long-term accessibility by tagging memories for sleep spindle-related reprocessing.
Sleep has been shown to promote the consolidation of newly acquired information actively (Born & Wilhelm, 2012). Slow oscillations and sleep spindles in particular have been repeatedly linked to the “offline” consolidation of newly stored memories (Marshall, Helgadottir, Molle, & Born, 2006; Schabus et al., 2004). However, recent results suggest that sleep does not uniformly consolidate encoded memory traces but rather that sleep “selects” or favors specific memories for consolidation.
It is assumed that such selection depends on salience tags, which become attached to new memories either during initial memory formation or after encoding as memory traces can be modified after learning by the specific intentions of the learner (such as memorizing a specific part of the learned material; for reviews, see Stickgold & Walker, 2013; Saletin & Walker, 2012). During memory formation, for example, sleep favors emotional over neutral stimuli (Nishida, Pearsall, Buckner, & Walker, 2008; Payne, Stickgold, Swanberg, & Kensinger, 2008; Hu, Stylos-Allan, & Walker, 2006) or memories cued to be remembered over memories cued to be forgotten (van Dongen, Thielen, Takashima, Barth, & Fernandez, 2012; Saletin, Goldstein, & Walker, 2011). After learning, later recall tests suggest that sleep favors memories bearing future relevance (Wilhelm et al., 2011) and those coupled with an expectation of reward (Fischer & Born, 2009).
Interestingly, selective consolidation of, for example, memories tagged to be of future relevance (compared with untagged memories) or memories cued to be remembered over memories cued to be forgotten has been shown to be specifically related to sleep patterns, such as sleep spindles and slow wave activity (Saletin et al., 2011; Wilhelm et al., 2011) as well as to the general time spent asleep (van Dongen et al., 2012). More specifically, memory performance of information cued to be of future relevance is positively correlated to slow wave activity, sleep spindle count, and fast spindle density over parietal regions as well as time spent asleep, with uncued memories not showing such correlations or even negative associations (van Dongen et al., 2012; Saletin et al., 2011; Wilhelm et al., 2011). A recent study by Hoedlmoser et al. (2015) also revealed robust negative relationships between sleep spindle activity (SpA; 12–15 Hz) during a diurnal nap and the forgetting of a gross motor skill. These results indicate not only that relevant information is consolidated by sleep but also that sleep might also actively instigate forgetting (Saletin & Walker, 2012).
However, surprisingly little is known concerning the neuronal correlates of these “tagged” memories over “untagged” during the memory formation process before sleep. Yet, this is of specific relevance as it would allow linking the neuronal signatures during memory formation to the sleep mechanisms involved in the consolidation process immediately thereafter. Moreover, the neuronal learning signature itself might allow a prediction of the performance levels the next day.
fMRI data indicate that the extent of hippocampal activity at initial memory formation (before sleep) relates to subsequent sleep-dependent consolidation (Rauchs et al., 2011). More specifically, in participants who were allowed to sleep after learning—but not in those who were prevented from sleeping—hippocampal responses during encoding were stronger for items correctly recognized after a delay of 3 days than for missed items. Hippocampal activity during encoding before sleep may therefore reflect “successful” tagging, which causes preferential or selective strengthening or replay of these memories during subsequent sleep.
Interestingly, in addition to predicting overnight memory consolidation, studies using the so-called “subsequent memory paradigm” have also linked hippocampal responses during initial memory encoding to subsequent memory performance after short retention periods in which the participants remain awake (for reviews, see Werkle-Bergner, Muller, Li, & Lindenberger, 2006; Paller & Wagner, 2002). In these studies, stimuli are post hoc divided into two categories: items that are recalled and items that are not recalled during a memory test. Contrasting neural responses of remembered versus not-remembered items highlight neural activity during memory formation that is predictive for subsequent memory access. Usually, neural levels are higher for subsequently remembered items.
These results suggest that there may be some overlap between neural tags reflecting successful memory formation and those that account for selective reactivation during sleep. Thus, studies on the subsequent memory effect appear to provide promising results not only for the investigation of successful memory formation but also to identify whether the degree of subsequent sleep-dependent consolidation is already determined during initial memory formation.
Pioneering studies investigating the subsequent memory effect used ERPs. Typically, items that are subsequently remembered elicit more positive ERPs in a latency range of 300–800 msec (for a review, see Werkle-Bergner et al., 2006). In addition to studies using ERPs, there is ample evidence linking EEG theta activity to successful encoding and/or synaptic plasticity with a postulated relation to the hippocampal formation (for reviews, see Nyhus & Curran, 2010; Hasselmo, 2005; Kahana, Seelig, & Madsen, 2001). Like fMRI studies that reveal stronger hippocampal activity during encoding for memories later remembered, EEG research consistently points to stronger theta synchronization during learning (Osipova et al., 2006; Sederberg, Kahana, Howard, Donner, & Madsen, 2003; Mölle, Marshall, Fehm, & Born, 2002; Klimesch, 1996, 1999). All these earlier studies assume that memory formation only occurs as participants study and encode information. However, there is compelling evidence from cognitive psychology that memory retrieval per se also contributes considerably to the formation and elaboration of memories (Karpicke & Roediger, 2008). Indeed, during memory retrieval, theta levels are known to be higher for remembered as opposed to nonremembered items (Klimesch, Doppelmayr, Schimke, & Ripper, 1997). EEG theta activity during memory formation before sleep might, therefore, like hippocampal activity in the fMRI, be “the” electrophysiological correlate reflecting the tag for later “offline” consolidation (during sleep).
Building on those findings, we sought to analyze oscillatory brain activity (1) during memory formation before sleep as well as (2) during subsequent night sleep. This would allow us to find oscillatory EEG correlates during memory formation, which not only predict recall performance the next day but also directly relate to specific sleep features that have been consistently linked to sleep-dependent memory consolidation.
The data used in this study were published by Schabus et al. (2004). The focus of the current study is different and includes a detailed analysis of oscillatory brain activity during memory recall before sleep.
The study was conducted in accordance with the ethical principles in the Declaration of Helsinki and was approved by the local research ethics committee (University of Salzburg). Before participation, all participants gave written informed consent.
Twenty-four participants (12 men) aged between 20 and 30 years (mean age = 24.42 years, SD = 2.59 years) participated in this study. All participants were students of medicine or psychology and fulfilled the following inclusion criteria: right handed, nonsmoker, no history of severe organic or mental illness, no signs of mood disorders (Self-Rated Anxiety Scale [raw score] < 36 [Zung, 1971], Self-Rated Depression Scale [raw score] < 40 [Zung, 1965]), and no sleep disturbances (Pittsburgh Sleep Quality Index Global Score < 5; [Buysse, Reynolds, Monk, Berman, & Kupfer, 1989]). Furthermore, we excluded sleep disorders (such as sleep apnea or periodic leg movements) by means of a screening polysomnography (PSG) at the start of the study. Participants' daily sleep quality and sleep–wake rhythm were controlled using sleep logs and wrist-worn actigraphs (Cambridge Neurotechnology Actiwatch, Cambridge, United Kingdom).
Each participant completed five laboratory visits. Visit 1 served as a general pre-examination for assessing participants' anamnesis as well as completing a number of psychometric tests. In Visits 2, 3, and 4, participants slept for a whole night in the laboratory with complete PSG. Visit 2 served diagnostic and acclimatization purposes, whereas Visits 3 and 4 were either preceded by a declarative word pair task (experimental PSG [EPSG]) or by a nonlearning control procedure (control PSG [CPSG]). Visits 3 and 4 were separated by 7 (±1) days, and the assignment of both tasks to Visits 3 and 4 was counterbalanced across all participants. One week after EPSG, a follow-up test of participants' memory performance was carried out (Visit 5).
All night sleep recordings started between 11:00 p.m. and midnight and were terminated after 8 hr in bed.
The word pair task consisted of 160 semantically unrelated word pairs. All word pairs were presented twice in randomized order ∼2.5 hr before sleep (Encoding). Both learning trials were separated by a short break of 2 min. Each word pair was presented on a computer screen for 6.5 sec with an ISI of 3.5 sec (to confine the usage of different mnemonic strategies, participants were instructed to imagine a relation between pairs of words visually).
After completing the encoding session, participants' immediate memory performance was tested using a cued recall procedure (without feedback) with cue words presented in a different order than in the learning trials (Recall 1). The first word of a pair was presented as a cue, and participants were asked to speak the corresponding target word within 6.5 sec. After 6.5 sec or after participants have pressed a response button, a gray fixation cross was presented for 3.5 sec during which participants were instructed to relax, stop attempting to retrieve the target word, and await the next cue word. The cued recall test was repeated in the morning approximately 30 min after waking (Recall 2; see Figure 1A) and again 1 week later (Recall 3).
The response score consisted of the number of correct responses and the number of (unambiguous) semantic correct answers (such as “flow” or “stream” instead of “river”), which were weighted by a factor 0.5. Test performance was expressed as the percentage of correct responses (e.g., Recall 1 = [correct response scoreRecall1 / 160] × 100). Consequently, the overnight change in declarative memory performance (overnight memory change [OMC]) was calculated by subtracting Recall 1 from Recall 2 performance. According to their OMC, participants were post hoc classified as memory improvers (I+, N = 14, OMC > 0) and memory nonimprovers (I−, N = 9, OMC ≤ 0). Note that, because of technical problems, one participant had to be omitted from further analyses.
The control task was designed to be as similar to the memory task as possible, but lacking any intentional learning component. In place of word pairs, the control task consisted of pseudo word pairs in which size and font (italic/nonitalic) of some letters were changed. In the control “encoding” procedure (control Encoding), participants were instructed to count silently such deviating letters in each pseudo word pair. Thus, visual input and task duration were matched with the memory task, but participants were distracted from processing the pseudo word pairs on a deep, semantic level. Furthermore, task difficulty, cognitive strain, and tiredness can be assumed to be comparable between the memory task and nonlearning control procedure (Gais, Molle, Helms, & Born, 2002).
In the control “recall” (control Recall) session, pseudo word pairs were presented once more, but in contrast to control encoding, participants were asked to indicate the sum of all deviant letters within each pseudo word pair by pressing a respective response button. Control “recall” performance (perceptual accuracy) is expressed as the percentage of correct button presses ([correct button presses / 160] × 100). As with the word pair task, performance was tested before (control Recall 1) and after (control Recall 2) sleep, and the change (here, overnight change in perceptual performance [OPPC]) from evening to morning was calculated. Note that, because of technical problems with the response pad, perceptual performance values of three participants could not be used for statistical analysis.
PSG was recorded using Synamps EEG amplifiers (NeuroScan, Inc., El Paso, TX). EEG data were acquired using 21 gold-plated silver electrodes (Fp1, Fpz, Fp2, F7, F3, Fz, F4, F8, T3, C3, Cz, C4, T4, T5, P3, Pz, P4, T6, O1, Oz, and O2 as well as A1 and A2 for later rereferencing) attached to participants' scalps according to a standard system (Jasper, 1958). In addition, a number of physiological channels were recorded: five EOG channels, one bipolar submental EMG channel, one bipolar electrocardiogram channel, and one bipolar respiratory channel (chest wall movements). During the adaptation/screening night (Visit 2), a different PSG setup with eight EEG, four EOG, one electrocardiogram, three EMG (submental and left/right tibialis), and four respiratory channels (nasal airflow, chest and abdominal wall movements, oxygen saturation) was acquired.
During recording, all signals were online referenced to electrode FCz, digitized with a sampling rate of 250 Hz, and amplified between 0.10 and 70 Hz with a 50-Hz notch filter. Impedances were kept below 5 kΩ.
EEG data were analyzed using own-built Matlab routines (MathWorks, Natick, MA) and EEGLAB toolbox (Delorme & Makeig, 2004).
Event-related Changes of Spectral Activity during Task Performance
Note that we concentrated our analysis on memory recall rather than encoding, as final memory testing was conducted directly before a full night of sleep and as memory recall itself is known to have especially strong modulatory effects on memory representations (Roediger & Butler, 2011).
To compute oscillatory activity during memory recall (Recall) or perceptual stimulus processing (control Recall), respectively, we used the event-related spectral perturbation (ERSP) method as implemented in EEGLAB. The ERSP measures the time course of a stimulus-induced change in spectral power of a predefined frequency range from preoccurrence to postoccurrence of an experimental stimulus (Delorme & Makeig, 2004), comparable with the classical event-related synchronization/desynchronization(ERS/ERD) method (Pfurtscheller & Aranibar, 1977). Note that positive values reflect an increase in oscillatory activity from before to after cue word appearance, whereas negative values reflect a decrease.
At first, EEG signals were rereferenced to the linked mastoids (A1, A2) and corrected for eye blinks and eye movements using independent component analysis. Thereafter, a thorough manual artifact inspection was performed to correct for remaining EOG, EMG, or movement artifacts. Subsequently, signals were band-pass filtered (1–30 Hz, zero phase shift) and segmented into epochs ranging from −1000 msec before to 2000 msec after cue word appearance. We included correctly remembered (HITS) and nonremembered (MISSES) cue words into our computations as both account for the total amount of OMC. Mean number of artifact-free epochs were 135.70 (SD = 13.89, range = 102–153) during Recall 1 and 145.48 (SD = 11.47, range = 113–160) during control Recall 1. Average event-related changes in spectral EEG power from before to after cue word appearance were calculated for a broad frequency range from 3 to 25 Hz. ERSP computation was performed using a Morlet wavelet transformation with the number of wavelet cycles slowly expanding with frequency (three cycles at 3 Hz, 12.5 cycles at 25 Hz, 50 frequency bins, 200 time bins [−440 to 1436 msec]). Time bins before cue word onset were chosen as baseline (−440 to −8 msec). Resulting ERSP values within the theta band (4.06–7.13 Hz) were averaged for three consecutive, post-cue word time windows (4–504, 504–1004, and 1004–1436 msec). Note, as the mean RT over conditions (Recall and control Recall) was 2184.22 msec (SD = 287.83 msec, range = 1433.13–2899.34 msec), we restricted our analysis to time windows before 1500 msec to minimize the probability (99.89% of mean RTs were over 1500 msec) of including motor artifacts because of button presses or time windows in which the memory search was completed much earlier.
In a subsequent step and to focus on theta power over electrodes reflecting memory recall (rather than unspecific perceptual stimulus processing), we computed statistical difference maps between Recall 1 and control Recall 1 theta ERSP separately for the I+ and I− groups as well as for the Group (I+ vs. I−) × Condition (control Recall 1 vs. Recall 1) interaction (see Figure 2). For correlational analyses, we used participants' mean difference between Recall 1 and control Recall 1 theta ERSP (Recall 1 − control Recall 1, in the following termed “recall-induced theta ERSP”; see Figure 1B). Positive values reflect a stronger event-related synchronization (or less pronounced desynchronization) in spectral theta power during Recall 1 as compared with during control Recall 1. Instead, negative values reflect a stronger event-related desynchronization (or less pronounced synchronization) in spectral theta power during Recall 1 as compared with during control Recall 1). As theta activity during 500–1000 msec after cue word appearance differed most strongly between memory groups, we will only report results regarding this time frame.
Sleep stage classification for every 30-sec epoch was done automatically and in accordance to standard criteria (Rechtschaffen & Kales, 1968) by the SOMNOLYZER 24×7 (The Siesta Group, Vienna, Austria) and verified manually by a sleep scoring expert.
Sleep spindles were detected automatically on central (C3, C4) electrodes rereferenced to contralateral mastoids. Spindle detection was performed using a previously published algorithm (Anderer et al., 2005). In the first step, EEG signals were band-pass filtered between 10 and 18 Hz using a zero-phase fourth-order Butterworth filter. Thereafter, the instantaneous amplitude of that signal was computed via Hilbert transformation. Finally, spindles were automatically detected according to following criteria: (1) minimal amplitude of 12 μV, (2) spindle duration between 0.3 and 2.0 sec, (3) a frequency range of 11–15 Hz, (4) spindle onset within a non-rapid eye movement epoch, and (5) no co-occurring muscle (30–40 Hz) and/or alpha (8–12 Hz) artifacts. For each detected sleep spindle, the SpA (Amplitude × Duration, μVs) was determined. Spindle events with a mean frequency of <13 Hz were classified as slow sleep spindles, and spindle events with a mean frequency of >13 Hz were classified as fast sleep spindles. As spindle results from electrode positions C3 and C4 strongly resembled one other, we chose to report results for only electrode C3 in the following.
To highlight memory task-induced changes in oscillatory activity during sleep, we computed the percent change of participants' mean SpA during the CPSG to the EPSG (see Figure 1C), which was termed “consolidation-related SpA.” Positive values reflect an increase in SpA from the CPSG to the EPSG; negative values reflect a decrease.
Statistical analyses were performed using Matlab's Statistics toolbox, EEGLAB's statistics toolkit, and PASW Statistics 18.0.2 software (SPSS, Inc., Chicago, IL). All statistics were calculated using two-tailed tests. Comparisons between groups were performed by independent samples t tests (I+ vs. I−), and comparisons within conditions (control Recall 1 vs. Recall 1 or CPSG vs. EPSG, respectively) were computed by paired-samples t tests. Group × Condition interactions were calculated using mixed within–between participants repeated-measure ANOVAs. For correlational analyses, we used two-tailed Pearson's bivariate correlations. Finally, to disentangle the triangle relationship between SpA during sleep, recall-induced theta activity before sleep, and overnight changes in memory performance, a mediation analysis using a bootstrapping (5000 bootstrap resamples) method with bias-corrected confidence estimates (95%) was carried out (Preacher & Hayes, 2004, 2008). Whenever results had to be corrected for multiple testing, significance thresholds were corrected using the false discovery rate (Benjamini & Yekutieli, 2001) with a threshold value of alpha = 0.1. In all other cases, significance threshold was set to 5%.
All performance levels are depicted in Table 1, separately for memory improvers (I+) and nonimprovers (I−). Note that absolute memory performance did not differ between I+ and I− at any test time (Recall 1, 2, and 3; all ps > .33). However, I+ and I− differed significantly in their OMC. Whereas I+ showed a significant overnight increase in memory performance (t(13) = −5.80, p < .001), I− significantly decreased their memory performance from presleep to postsleep (t(8) = 4.35, p = .002). Regarding control recall, I+ and I− differed neither during evening control recall (control Recall 1), morning control recall (control Recall 2), nor in their OPPC.
|.||I+ .||I− .||Student's t Test .|
|Mean .||SE .||Mean .||SE .||t .||p .|
|Control Recall 1||88.98||3.71||89.51||1.77||0.12||.91|
|Control Recall 2||92.24||2.37||92.92||1.63||0.22||.83|
|.||I+ .||I− .||Student's t Test .|
|Mean .||SE .||Mean .||SE .||t .||p .|
|Control Recall 1||88.98||3.71||89.51||1.77||0.12||.91|
|Control Recall 2||92.24||2.37||92.92||1.63||0.22||.83|
Independent-samples t tests depict the differences between I+ (memory improver) and I− (memory nonimprover) at all memory tests (Recalls 1, 2, and 3). Note that memory groups only differ in their OMC but not in their absolute memory or control recall performance (control Recalls 1 and 2) nor in their OPPC. All data are presented in percent.
Furthermore, I+ and I− groups did not differ in other (possibly confounding) variables, like fatigue (as controlled by 100-mm visual analog scales, all ps > .3), age (I+: mean = 25.00, SD = 2.54; I−: mean = 24.00, SD = 2.51; p > .3), IQ (I+: mean = 114.14, SD = 11.31; I−: mean = 116.67, SD = 15.81; p > .6), general memory capacity (I+: mean = 114.43, SD = 9.23; I−: mean = 118.00, SD = 18.44; p > .5), and gender distribution (I+: eight men, six women; I−: four men, five women; p = .68).
Theta ERSP during Memory Formation before Sleep
A mixed-model repeated-measure ANOVA with the dependent variable theta ERSP revealed a significant Group (I+ vs. I−) × Condition (control Recall 1 vs. Recall 1) interaction over a left fronto-centroparietal electrode cluster (F3, Cz, P3, Pz, and P4; all ps < .019). Post hoc paired t tests indicate a significant stronger theta ERSP during Recall 1 compared with control Recall 1 for I+ over almost all electrode sites (Fp1, Fpz, Fp2, F7, F3, Fz, F4, Cz, P3, Pz, and P4; all ps < .038), although the effect was most pronounced over left frontal regions. In contrast, I− showed similar theta responses during the control and experimental conditions. Also note that the groups (I+, I−) only differed in Recall 1 theta ERSP (Fp1, Fpz, Fp2, F7, F3, T3, C3, Cz, T5, P3, Pz, P4, T6, O1, Oz, and O2; all ps < .075) but not in control Recall 1 theta ERSP (all ps > .061). This indicates that revealed differences are related to state-specific differences during declarative memory recall in the evening rather than unspecific (trait-like) differences between participant groups (see Figure 2).
Furthermore, to test for a linear relationship between recall-induced theta ERSP (Recall 1 − control Recall 1) and OMC, Pearson correlations were computed. As the described Group (I+ vs. I−) × Condition (control Recall 1 vs. Recall 1) interaction was most pronounced over electrode position Pz (F(1, 21) = 11.78, p = .0025), correlational analyses were primarily focused on this parietal region. As depicted in Figure 3A, positive values in recall-induced theta ERSP relate to an overnight increase in memory performance, whereas negative values relate to a decrease in memory performance from Recall 1 to Recall 2 (r23 = .643, p < .001). In addition, to investigate whether this relationship is restricted to electrode position Pz, we computed topographical maps depicting the strength (Pearson's r) and p values over the scalp. As shown in Figure 3B, this relationship was prominent over a widespread left fronto-centroparietal electrode cluster (F3, Fz, C3, Cz, P3, Pz, and P4; all ps < .041), which is strongly overlapping with the cluster revealing the strongest recall-induced theta differences between memory groups (F3, Cz, P3, Pz, and P4).
In addition, to test whether recall-induced theta ERSP during Recall 1 also relates to more long-lasting changes in individual memory performance, another Pearson correlation was calculated using recall-induced theta ERSP (Pz) and performance change from Recall 1 to the delayed recall after 1 week (Recall 3). Interestingly, also performance changes over 1 week show a positive relationship to recall-induced theta ERSP and just fail to reach significance (r23 = .41, p = .05). These data imply that recall-induced theta synchronization relates not only to overnight performance enhancement but also to less time-dependent forgetting over the duration of 1 week.
Moreover, we want to stress the specificity of that effect. Please note that recall-induced theta ERSP during Recall 1 was only related to relative changes in memory performance (overnight and over a period of 1 week) rather than to participants' absolute memory performance (Recall 1: r23 = −.27, p = .22; Recall 2: r23 = −.19, p = .4; Recall 3: r23 = −.07, p = .75) or their general memory capacity (r23 = .10, p = .66).
Consolidation-related Sleep SpA (Data Reproduced from Schabus et al., 2004)
As has been shown (Schabus et al., 2004), consolidation-related changes in fast sleep SpA are positively correlated to overnight changes in memory performance (r23 = .53, p < .01). Participants showing an increase in fast SpA in the night after the memory task, as compared with the night after performing the nonlearning control procedure, benefit more strongly from sleep (see Figure 4). Note that computing the same correlation using changes in slow SpA did not lead to significant results (r23 = .24, p = .27). Also note that results are not identical with those reported by Schabus and colleagues (2004); as in this study, we included all non-rapid eye movement spindle detections into our analysis rather than N2 spindles only, and we adapted filter settings and detection criteria (for details, refer to Schabus et al., 2004).
Addressing the “Big Picture”: A Hypothesized Triangular Relationship
Our results revealed a relationship between overnight changes in declarative memory performance and (1) recall-induced theta ERSP as well as (2) consolidation-related changes in fast sleep SpA. To determine whether recall-induced theta ERSP also directly relates to consolidation-related changes in fast sleep SpA and whether the relation between fast sleep SpA and OMC is mediated by recall-induced theta ERSP before sleep, a mediation analysis was conducted.
The mediation analysis indicates that recall-induced theta ERSP is positively related to changes in fast sleep SpA (B = 0.078, t(21) = 2.68, p = .014; Figure 5, left). Furthermore, the analysis reveals that, even when controlling for the contribution of fast sleep SpA, the positive association between theta responses before sleep and overnight change in memory performance remains significant (B = 1.415, t(21) = 2.67, p = .015; Figure 5, right). Finally and most interestingly, results confirm a mediating role of recall-induced theta ERSP before sleep for the relation between SpA and OMC (B = 0.106, 95% CI [0.038, 0.226]). Even more surprisingly, the results indicate that the relation between fast SpA and OMC disappears when controlling for theta responses during initial memory formation (B = 0.119, t(21) = 1.45, p = .162; Figure 5, bottom). This reveals a neglected variable (theta ERSP during memory formation) and a possible “triangular relationship.”
This study addressed the role of spectral theta power during memory formation in the evening before sleep and its influence on sleep-dependent memory consolidation. Specifically, we expected that theta ERSP during memory formation predicts OMCs and does so by driving sleep spindle-related memory replay during the night.
Data revealed that participants with an overnight improvement in declarative memory performance did display stronger recall-induced theta synchronization during initial memory formation in the evening and also stronger consolidation-related increases in fast sleep SpA. Furthermore, in line with our expectations, presleep theta ERSP was directly linked to consolidation-related changes in fast SpA, indicating that processes during memory formation before sleep might indeed influence subsequent spontaneous consolidation processes during sleep. Interestingly, it can be speculated that the reported relationship between sleep spindles and overnight performance changes (Schabus et al., 2004, 2008; Clemens, Fabo, & Halasz, 2005; Gais, Helms, et al., 2002) are significantly influenced by theta oscillations during initial memory formation. Thus, spectral theta power during memory formation before sleep might be a missing element for understanding the relationship between OMCs and postlearning sleep activity.
Specifically, we found that an increase in fast (but not slow) sleep SpA during the night after declarative memory formation—as compared with after a nonlearning control procedure—is positively related to the amount of overnight changes in memory performance. This corresponds to findings relating neural firing patterns of sleep spindles to enhanced synaptic plasticity (Rosanova & Ulrich, 2005) as well as data indicating that fast (but not slow) sleep spindles occur preferentially within up-states of cortical slow oscillation, a time frame linked to the transfer of initial hippocampal to long-term cortical memory representations (Born & Wilhelm, 2012). Altogether, an increase in SpA might thus very well reflect a more efficient consolidation of freshly encoded memories before sleep.
In this context, it is also important to note that for analyzing (1) changes in spectral theta power at memory formation as well as (2) changes in SpA during sleep, we here rely on difference measures between a declarative word pair task and a nonlearning control procedure. This allows for the important control for trait-like differences between participants influencing memory performance such as lower tonic theta power (Klimesch, Vogt, & Doppelmayr, 2000) and higher (fast) SpA in highly gifted individuals (Schabus et al., 2008).
Moreover, we found differences in recall-induced theta ERSP between memory improvement groups (I+ and I−). As we did not find such differences for the nonlearning control task, these can be linked directly to processes reflecting memory formation rather than general activity patterns to visual stimuli. More specifically, only memory improvers showed stronger theta responses during declarative memory recall compared with the nonlearning control procedure. In contrast, theta responses of memory nonimprovers were not altered during memory recall. In line with these findings, recall-induced theta ERSP before sleep showed a strong linear relationship to OMC with increments in theta ERSP relating to an increase and decrements relating to a decrease in memory performance. Interestingly, recall-induced theta synchronization is also positively correlated with performance after a week delay period. Thus, increments in recall-induced theta ERSP do not only relate to an overnight increase in declarative memory performance but also to less time-dependent forgetting over a period of 1 week.
Overall, as our data indicate (1) that memory groups (improvers vs. nonimprovers) do not differ in their absolute memory performance before sleep and as (2) recall-induced theta ERSP is not correlated with participants' general memory aptitude, we believe that theta differences before sleep are indicating immediate state-driven changes overnight rather than trait-like differences in general memory capacity.
These theta oscillations have been shown to play a crucial role in human episodic memory (Nyhus & Curran, 2010; Klimesch, 1999), and it has been proposed that theta oscillations provide top–down control to the hippocampus to influence the retrieval of newly acquired episodic memories. More precisely, it is suggested that the interplay between theta and gamma oscillations during memory recall or recall attempts reflect a cortico-hippocampal interaction, which ultimately lead to the simultaneous activation and reinstatement of all features of a memory representation in the cortex (Nyhus & Curran, 2010). On the basis of these findings, we speculate that stronger recall-induced theta responses in participants, who improve their memory performance overnight, reflect a more efficient cortico-hippocampal cross-talk and thereby further refine or “tag” the newly acquired memory representations for subsequent reprocessing during sleep.
The connection between theta responses before sleep and changes in fast SpA during sleep revealed by our results support such an interpretation (cf. Figure 5). This suggests that theta oscillations are not only influencing memories' long-term accessibility but also may trigger sleep mechanisms required for overnight consolidation.
Recent studies suggest that sleep is not beneficial for all memories but rather selects specific memories for consolidation (van Dongen et al., 2012; Rauchs et al., 2011; Saletin et al., 2011; Wilhelm et al., 2011; Fischer & Born, 2009; Payne et al., 2008). It can be assumed that whether sleep is contributing to consolidation is based on salience tags attached to memory traces (Stickgold & Walker, 2013). Recall-induced theta ERSP may thus reflect presleep memory tagging that is crucially influencing subsequent sleep-dependent memory consolidation. Interestingly, our mediation analysis suggests that the relation between sleep spindle-related consolidation and OMC is not as direct as formerly believed by us and others. Rather, it appears to be initiated or mediated via earlier processes that already begin during memory formation in the form of theta oscillations.
In conclusion, our study addresses a much neglected fragment in the sleep-dependent memory consolidation hypothesis, namely, the initial memory formation process. Findings reveal that memory tagging before sleep can be studied by focusing on theta brain oscillations and that these may even determine whether a memory trace will be strengthened across a period of sleep. Specifically, we suggest that recall-induced theta ERSP during initial memory formation reflects cortico-hippocampal interactions and leads to better long-term accessibility by “tagging” memories for sleep spindle-related reprocessing.
This work was supported by the Austrian Science Fund (FWF) Project I-934-B23. In addition, D. P. J. Heib is associated with and financially supported by the Doctoral College “Imaging the Mind” (F. W. F.; W1233). We thank Cornelia Sauter (Department of Neurology, Vienna), Gerhard Klösch (Department of Neurology, Vienna), Silvia Parapatics (Department of Psychiatry, Vienna), Bernd Saletu (Department of Psychiatry, Vienna), and Waltraud Stadler (Department of Sports and Health Science, Munich) for the contribution on the earlier study.
Reprint requests should be sent to Manuel Schabus, Laboratory for Sleep, Cognition and Consciousness Research, Department of Psychology, University of Salzburg, Hellbrunnerstrasse 34, 5020 Salzburg, Austria, or via e-mail: Manuel.Schabus@sbg.ac.at.