Bayesian meta-analysis reveals the mechanistic role of slow oscillation-spindle coupling in sleep-dependent memory consolidation

eLife Assessment

This important study presents a meta-analysis confirming a statistically significant association between slow oscillation-spindle coupling and memory formation, although the reported effects are limited (~0.5% of variance). The evidence is overall convincing, but the statistical methods may be difficult to follow for readers unfamiliar with advanced techniques. This work will be of particular interest to neuroscientists studying the neural mechanisms of sleep and memory.

https://doi.org/10.7554/eLife.101992.3.sa0

Significance of the findings:

Important: Findings that have theoretical or practical implications beyond a single subfield

Landmark
Fundamental
Important
Valuable
Useful

Strength of evidence:

Convincing: Appropriate and validated methodology in line with current state-of-the-art

Exceptional
Compelling
Convincing
Solid
Incomplete
Inadequate

During the peer-review process the editor and reviewers write an eLife Assessment that summarises the significance of the findings reported in the article (on a scale ranging from landmark to useful) and the strength of the evidence (on a scale ranging from exceptional to inadequate). Learn more about eLife Assessments

Abstract
Introduction
Results
Discussion
Methods
Appendix 1
Appendix 2
Appendix 3
Appendix 4
Appendix 5
Appendix 6
Data availability
References
Article and author information
Metrics

Abstract

The active system consolidation theory suggests that information transfer between the hippocampus and cortex during sleep underlies memory consolidation in humans. Neural oscillations during sleep, including the temporal coupling between slow oscillations (SO) and sleep spindles (SP), may play a mechanistic role in memory consolidation. However, differences in analytical approaches and the presence of physiological and behavioral moderators have led to inconsistent conclusions. This meta-analysis, comprising 23 studies and 297 effect sizes, focused on four standard phase-amplitude coupling measures including coupling phase, strength, percentage, and SP amplitude, and their relationship with memory retention. We developed a standardized approach to incorporate non-normal circular-linear correlations. We found strong evidence supporting that precise and strong SO-fast SP coupling in the frontal lobe predicts memory consolidation. The strength of this association is mediated by memory type, aging, and spatiotemporal features, including SP frequency and cortical topography. In conclusion, SO-fast SP coupling should be considered as a general physiological mechanism for memory consolidation.

Introduction

Over the past three decades, accumulating evidence supports the role of sleep neural oscillations and their cross-frequency coupling in the spatiotemporal coordination across brain regions, supporting sleep-dependent memory consolidation (Buzsáki and Draguhn, 2004; Hyafil et al., 2015; Klinzing et al., 2019). During nREM sleep, oscillatory activities including cortical slow oscillations (SO; 0.16–4 Hz, Figure 1A), thalamocortical sleep spindles (SP; 8–16 Hz), and hippocampal sharp-wave ripples (SWR; 80–300 Hz), along with their long-distance coordination, are widely believed to be closely associated with the process of transferring temporary encoded memory traces to cortical networks for consolidation (Clemens et al., 2007; Maingret et al., 2016; Staresina et al., 2015).

Figure 1

Download asset Open asset

Measurement of phase-amplitude coupling (PAC) in slow oscillation and spindle events.

(A) The origin of neural oscillations during sleep. SO slow oscillation, *SP sleep spindle, SWR* sharp wave ripple. In each subgraph, the vertical line indicates the typical amplitude of that sleep wave, while the horizontal line indicates the typical frequency and duration. Note that SPs also propagate along the cortex, and the figure only displays the origin. (B) Electrophysiology representation diagram of the SO-SP coupling. SO and SP amplitudes are normalized. The phase of SOs when SPs are at their maximum instantaneous amplitude is recorded as the coupling phase. The occurrence of SPs and SO-SP coupling is not necessarily continuous as shown in the diagram. (C) Coupling preferred phase and strength diagram. (Left) The phase and strength used in the circular plot are simulated data from existing dataset for visualization purposes only. At the group level, the mean circular direction shows the preferred SO phase, while the mean vector length shows the strength of the precise coupling. (Right) The phase of SO peaks is noted as 0, while the phase of SO troughs is noted as $\pm π$ .

Given that non-invasive recordings (e.g.; EEG) cannot accurately detect SWRs in deep brain structures, electrophysiological studies in humans primarily focus on the role of SOs and SPs in memory consolidation. Consistent evidence indicates that, following intensive learning, SP density and its co-occurrence with SOs significantly increase compared to baseline nights and control groups (Gais et al., 2002; Mölle et al., 2009; Mölle et al., 2004; Schmidt et al., 2006; Solano et al., 2022). Correspondingly, measures of SPs and their coupling with SOs predict over-sleep retention performance of newly acquired memories (Holz et al., 2012; Nicolas et al., 2022; Kumral et al., 2023; Kurdziel et al., 2013; Rodheim et al., 2023).

The role of sleep oscillations in memory consolidation is supported by consistent neurobiological foundations. SOs are generated and dominant in the prefrontal cortex during slow-wave sleep (SWS) and propagate anteriorly to posteriorly as traveling waves (Massimini et al., 2004; Kurth et al., 2017; Malerba et al., 2019; Niethard et al., 2018). The corticothalamic input of SOs during their depolarization up-state phase organizes the synchronous occurrence of SPs in the thalamic reticular nucleus. Subsequently, SPs propagate back widely to cortical areas through synchronized thalamocortical projections, resulting in EEG-measured SPs (Contreras et al., 1997; Kim et al., 1995; Marshall et al., 2006; Neske, 2015; Oyanedel et al., 2020; Steriade, 2003). The peak discharge of SPs in the cortex is associated with increased dendritic Ca²⁺ synchronization (Niethard et al., 2018; Rosanova and Ulrich, 2005; Seibt et al., 2017). This precise association enhances synaptic plasticity, leading to long-term changes in synaptic connections between cortical neurons, a mechanism for memory consolidation (Lindemann et al., 2016; Martin et al., 2000; Miyamoto et al., 2017; Steriade and Timofeev, 2003).

Despite extensive research on the functions of these oscillations during sleep and their causal sequences, it is unclear how the spatiotemporal coordination mechanisms facilitate specific temporal sequences of the peak discharge and the associated memory consolidation. Neuronal activity exhibits various forms of cross-frequency coupling, mainly including phase-phase coupling (PPC) and phase-amplitude coupling (PAC). The coupling is considered crucial in regulating the integration of information across multiple spatial and temporal scales (Canolty and Knight, 2010; Palva et al., 2005). Studies related to memory often focus on the unidirectional PAC regarding the strong amplitude modulation of faster oscillations (FO) driven by the phase modulation of relatively slower oscillations in a hierarchical order. The phase of SOs influences rhythmic spikes of FOs near the up-state peak or down-state trough of SOs (Canolty et al., 2006; Fell and Axmacher, 2011; Jensen and Colgin, 2007). In memory networks, PAC is considered a crucial mechanism supporting neural communication and plasticity (Fell and Axmacher, 2011).

The active system consolidation theory supports the subdivision of memory consolidation processes into three coordinated steps, which occur in the precise hierarchical temporal structure of SO-FO coupling involving SOs, SPs, and SWRs (Born and Wilhelm, 2012; Takehara‐Nishiuchi, 2021; Winocur and Moscovitch, 2011): (1) Hippocampal SWRs driven by cortical SOs facilitate the repeated replay of newly encoded memories. (2) The maximum amplitude of SWRs precisely locks with the trough phase of SPs during the depolarized up-state of SOs (SP-SWR coupling). (3) The maximum amplitude of SPs couples with the up-state of SOs (SO-SP coupling). These phase-locked coupling dynamics are considered integral to consistent communication between the hippocampus and neocortex, constituting a key mechanism for the reactivation and redistribution of temporary memory traces across cortical areas (Diekelmann and Born, 2010; Helfrich et al., 2019).

Pharmacological and stimulation research has provided evidence for causal relationships between coupling and memory consolidation. For example, the increasing level of inhibitory neurotransmission driven by GABAergic drug zolpidem improves the phase precision and strength of SO-SP coupling and, in turn, contributes to enhanced memory retention performance (Carbone et al., 2021; Kersanté et al., 2023; Niknazar et al., 2015; Zhang et al., 2020). Studies using calcium imaging and SP stimulation explained the significance of the precise coupling phase for synaptic plasticity: SP spike discharges extracted through electrical stimulation during SO up-states efficiently modify excitatory neocortical synapses (Rosanova and Ulrich, 2005). Recently, Niethard et al., 2018 observed that only SP spikes occurring around upstate peaks of SOs were accompanied by amplified calcium activity patterns to optimize synaptic plasticity. This spatiotemporal mechanism represents a critical perspective in the study of SO-SP coupling in memory consolidation.

When quantifying cross-frequency coupling, the spike-timing-dependent transfer theory (Diekelmann and Born, 2010; Rasch and Born, 2013) signified the importance of coupling phases and inconsistency in FO amplitudes. The coupling phase has been widely shown to reflect dynamic functional configurations, indicating remote communication and precise timing changes in synaptic activity among neural populations sharing cognitive functions (Jiang et al., 2015; Sauseng and Klimesch, 2008). Comparing to the depolarization state (positive half-wave) of cortical SOs which drives FOs including SPs, its inhibitory hyperpolarization state exhibits relative silence of neurons (Neske, 2015; Mölle et al., 2002). The preferred phase represents the phase of SOs at the maximum amplitude of FOs (Figure 1B and C). Another common method is the peri-event time histogram (PETH), representing the proportion of FOs centered (i.e. maximum trough) at different SO phase bins (Kurz et al., 2021; Muehlroth et al., 2019).

Phase alone does not reveal the inconsistent strength of FO amplitude between coupled and non-coupled SO phases within each time window. Therefore, the analysis of phase-amplitude distribution constitutes the so-called coupling strength, also defined as coupling inconsistency. In memory research, two commonly used quantification methods for coupling strength include mean vector length (Canolty et al., 2006; MVL) and modulation index (Tort et al., 2008; MI). Both determine the strength of PAC by measuring the uneven distribution of FO amplitude in different SO phase directions. Past simulation studies have proved the effectiveness of these two methods in extracting coupling strength under different conditions and the existence of noise (Hülsemann et al., 2019; Samiee and Baillet, 2017; Tort et al., 2010).

Compared to the coupling phase and strength, which reflect the contrast on the phase-amplitude level, recent research also focuses on the overall occurrence of coupling events across the night. This involves assessing the prevalence of co-occurrence, including the coupling percentage (Denis et al., 2021; Hahn et al., 2020; Halonen et al., 2021; Kurz et al., 2021) (i.e. co-occurrence rate, % coupling events / oscillation events), coupling counts (Niknazar et al., 2015), and the coupling density (Mylonas et al., 2020). It is reasonable to categorize all measures related to co-occurrence under a category defined as ‘coupling prevalence’, in contrast to the phase and strength (see Figure 2).

Figure 2

Download asset Open asset

Hierarchical diagram of coupling measures.

*PETH* peri-event time histogram, *MVL* mean vector length, MI modulation index. *SPcSO* Percentage of SPs coupled with SOs in all SP events. In contrast to the term ‘frequency’ used throughout the text in reference to the neural oscillation, the coupling prevalence in the diagram indicates the occurrence of SO–SP coupling events.

In SO-SP coupling studies, although most of them supported the importance of coupling phase in predicting memory retention, this association varies in magnitude across memory types, sleep stages, and age (Bastian et al., 2022; Cox et al., 2018; Denis et al., 2021; Donnelly et al., 2022; Hahn et al., 2020; Halonen et al., 2021; Halonen et al., 2022; Helfrich et al., 2018; Kurz et al., 2021; Mikutta et al., 2019; Weiner et al., 2024). In contrast, the significance of coupling strength and percentage on memory consolidation is inconsistent across studies (Kurz et al., 2023; Mikutta et al., 2019; Weiner et al., 2024; Hahn et al., 2022).

Furthermore, previous meta-analyses of SP function consistently indicate that SP amplitude (and power) is the most effective predictor of memory consolidation and cognitive abilities (Kumral et al., 2023; Ujma, 2021). The measure of neural oscillation amplitude is reported based on changes compared to mean amplitude, and the absolute amplitude difference between peak and trough reflects the intensity of synchronized neural activity near that specific scalp region (Teplan, 2002). Although SP amplitude is not a direct measure of SO–SP coupling, and prior studies report discrepancies regarding whether its maximum value predicts coupling phase and strength (Baena et al., 2023; Roebber et al., 2022), evidence consistently indicates that the magnitude of SP amplitude is systematically modulated by SO phase, and this phase-dependent modulation constitutes a core mechanism of coupling (Helfrich et al., 2018; Klinzing et al., 2019; Staresina et al., 2015). Including SP amplitude reflects group differences in SP activity overnight and allows comparison of the roles of coupling metrics and SP alone, as examined in many of the included studies (Kurz et al., 2021; Niknazar et al., 2015; Ladenbauer et al., 2021). Also, only 4 studies reported SP amplitude separately for coupled events, which limits targeted analyses. Therefore, we also include the mean peak-to-trough amplitude and power of all SP events in the meta-analysis (Figure 1B).

Electrophysiology evidence from studies included in our meta-analysis (Denis et al., 2021; Hahn et al., 2020; Mylonas et al., 2020) and others (Rodheim et al., 2023; Muehlroth et al., 2019; Bartsch et al., 2019) reported that the association between memory consolidation and SO-SP coupling is influenced by a variety of behavioral and physiological factors under different conditions. Among the moderators that have received significant attention in recent research are memory types, development and aging, pharmacological manipulations, disorders, and sleep stages. The early dual-process hypothesis proposed a dissociation between hippocampus-dependent and non-dependent memories reinforced during SWS and REM periods, respectively (Rasch and Born, 2013): The standard active consolidation theory supported the role of hippocampal-cortical nesting during nREM sleep for the consolidation of hippocampus-dependent memories, including verbal, visual, and spatial declarative memories that are episodic-related (Rasch and Born, 2013; Gais and Born, 2004; Marshall and Born, 2007). In contrast, some research has linked the processing of non-hippocampus-dependent memory, such as emotional and procedural memories, with REM sleep (Groch et al., 2013; Plihal and Born, 1997; Sara, 2017; Wagner et al., 2001).

This outdated theory may have led early SO-SP coupling studies to focus primarily on the experimental design of declarative memory inferences. Recent updates to this foundational theory propose that the hippocampus plays a crucial role in the consolidation of non-hippocampus-dependent memories during nREM sleep through the reactivation of spatiotemporal contextual features (Ackermann and Rasch, 2014; Boutin and Doyon, 2020; King et al., 2017; Sawangjit et al., 2018; Spencer et al., 2006). This has sparked discussions and research on whether hippocampal involvement in SO-SP-SWR coupling constitutes a general physiological principle in all types of memory, or at least in memories with spatiotemporal contexts. Particularly in the association between SO-SP coupling and non-hippocampal-dependent consolidation, recent studies have obtained conflicting results (Solano et al., 2022; Nicolas et al., 2022; Kumral et al., 2023; Cox et al., 2018; Mikutta et al., 2019; Mylonas et al., 2020; Hahn et al., 2022; Nishida and Walker, 2007; Wei et al., 2018).

In addition, early behavioral experiments have demonstrated the sensitivity of sleep-dependent memory consolidation to age-related changes in the brain, from development to aging (Kopasz et al., 2010; Spencer et al., 2007). The most significant changes of sleep neural oscillations during development include the maturation and dominance of frontal-originated global SOs and frontal-detected SPs (Hahn et al., 2019; Shinomiya et al., 1999; Timofeev et al., 2020). This change is associated with the improvement of memory consolidation from childhood to adolescence. In older adults, the time spent in SWS sharply decreases, followed by a reduction in the number and amplitude of SOs (Harand et al., 2012; Petit et al., 2004). Their prefrontal SP events decrease by over 40% compared to young adults and are correlated with their weakened memory performance (Mander et al., 2013; Martin et al., 2013).

Taking a step further, Helfrich et al., 2018 and Hahn et al., 2020 have respectively identified the modulating roles of aging and development in the frontal SO-SP coupling, including changes in the precision of coupling phase, shifts in coupling topography, as well as improvements and impairments in memory retention performance. Without exception, all age-related studies on sleep-dependent memory consolidation emphasize the involvement of the prefrontal cortex and posterior hippocampus, which contribute to episodic memory processing and consolidation. Differences in their gray matter volumes, and changes in structural integrity during development and aging, are related to the representation of neural oscillations (Mander et al., 2013; Ishii et al., 2018; Saletin et al., 2013).

Besides these moderators, overlooked in most studies is the region-specificity of different memory types and the frequency of SPs (Geva-Sagiv and Nir, 2019). The topography and frequency distribution coupling are neglected or subject to misleading interpretations when correlating other measures. SP frequency has traditionally been defined between 12 and 16 Hz (Rechtschaffen, 1968). However, recent research has found that SPs of different frequencies dominate at different phases of the SO cycle and in distinct cortical areas (Mölle et al., 2011a; Zeitlhofer et al., 1997). Therefore, studies started to analyze the SO-SP coupling by splitting SPs into fast and slow subtypes (Solano et al., 2022; Cox et al., 2018; Perrault et al., 2019). Since some studies found that fast SPs predominate in the centroparietal region, while slow SPs are more common in the frontal region, a significant amount of studies selectively extracted specific types of SPs from limited electrodes (Perrault et al., 2019; Schreiner et al., 2021; Dehnavi et al., 2021). Some studies even averaged all electrodes in their spectral and/or time-series analysis to estimate metrics of oscillations and their couplings (Nicolas et al., 2022; Mölle and Born, 2011b; Denis et al., 2022). This narrowed measure falls into the pitfall of considering all SPs as the regional oscillation and overlooking the out-of-phase distribution of SOs in different cortical areas (Nir et al., 2011).

Acknowledging the existence of regional SO and SP events, recent evidence supports that global SO and SPs, as traveling waves, constitute the majority of their oscillation cycles (Massimini et al., 2004; Dickey et al., 2021; Muller et al., 2016). Traveling waves are considered to play a crucial role in the transmission of information, including memory, between specific brain areas (Muller et al., 2018). Once projected onto the cortex, global SPs exhibit rotational characteristics, looping from the temporal lobe to the parietal and frontal lobes sequentially (TPF) and then travel back to the temporal lobe (Muller et al., 2016; O’Reilly and Nielsen, 2014). This results in phase gradients in the burst of SP spikes across different cortical areas, which represents the direction of asymmetric cortical propagation and might explain the phase shifts in coupling (Ermentrout and Kleinfeld, 2001; Hindriks et al., 2014). It is worth noting that in other types of oscillations, the traveling waves have been found to also follow descending frequency gradients (Zhang et al., 2018).

Dynamic spatiotemporal features of global SPs are believed to create necessary conditions for the synaptic plasticity (Dickey et al., 2021). Recently, relevant clues have emerged in research on SO-SP coupling. Hahn et al., 2020 and Helfrich et al., 2018 found that only the phase and strength of coupling detected in frontal electrodes had sufficient predictive power for the success of memory retention. Denis et al., 2021 reported a phase gradient from anterior to posterior area in SO-fast SP coupling by channels. Although all this evidence indicates the necessity for investigation of the spatiotemporal specificity of SO-SP coupling, current between-study differences and the ambiguity in defining SP types (see Discussion: Challenges of current statistical approaches in measuring EEG-behavior associations) introduce uncertainty to the measure of coupling topography. To our knowledge, there is no previous study that has systematically analyzed the dynamic spatiotemporal distribution of SO-SP coupling and its association with the frequency of oscillations and memory consolidation.

In summary, current research on the SO-SP coupling and memory consolidation faces multiple methodological challenges. Disadvantages of polysomnographic (PSG) studies include long recording times, limitations on subject recruitment, and variations in signal processing approaches. Researchers have the flexibility to choose analytical methods and selectively report results that favor their hypotheses (Cordi and Rasch, 2021; Reverberi et al., 2020), thereby increasing the risk of false positives and lack of reproducibility. Considering methodological limitations, varied analytical approaches, disparate quality of reports, and conflicting evidence, there is a clear urgency for a meta-analysis to consolidate the true effect sizes reported in studies, as well as propose standardized and targeted research and analysis methods.

To incorporate a broader range of studies without increasing between-study heterogeneity and publication bias, we requested unreported or selectively omitted effect sizes, as well as individual-level coupling data that were pre-processed but not included in correlation analysis. This approach helped us include more studies adopting a similar experimental design but using incomparable analysis methods, which is common in memory research. In addition, these data provide us with sufficient statistical power to compare spatiotemporal shifts in coupling phase and its association with memory consolidation.

The purpose of the current meta-analysis and review is to aggregate studies of SO-SP coupling to understand inconsistent results across studies. To be more specific, we used Bayesian hierarchical models to examine whether the cross-frequency coupling between SOs and SPs is associated with memory consolidation and determine which specific measure(s) of SO-SP coupling most accurately predict memory retention performance. In addition, we conducted moderator analyses to understand if physiological and behavioral factors modulate the strength of the relationship between SO-SP coupling and memory consolidation. We propose five questions for moderators: (1) Is SO-SP coupling a general physiological mechanism for memory consolidation or targeted at specific type(s) of memory?; (2) How do development and aging affect the association between coupling and memory consolidation?; (3) Is the coupling detected under specific (functional) region(s) more indicative of the memory retention?; (4) In which frequency range(s) of SPs is coupling more likely to be linked to memory consolidation?; (5) Does the stage and bout of sleep affect the measure of coupling-memory association? Besides the main study, we also analyzed the preferred distribution of coupling phases under different cortical areas and SP frequencies. In the rest of the discussion, we focus on summarizing conflicts and misinterpretations in current research, proposing strategies for research standardization, as well as discussing the implications of our results for future research focus and clinical applications.

Results

Study characteristics

In the final dataset of 23 studies included in the data analysis, the participants had an average age of 24.8 years, ranging from 7.3 to 78.0. The average female representation among participants was 49.5%. The average sample size for each study was 31.7 participants, ranging from 10 to 151. Each study contributed four effect sizes for each coupling measure on average. The included studies consist of 17 overnight and 6 nap studies, in which a total of 19 tasks for declarative memory and 6 tasks for procedural memory were measured. In addition, we assessed the risk of bias for each study before conducting data analysis (see Figure 3).

Figure 3

Download asset Open asset

Risk of bias assessment summary plot adapted from ROBINS-I (Sterne et al., 2016).

The most significant heterogeneity is revealed in the measurement of outcome, while the overall assessment indicated a moderate risk of bias across studies after requesting unreported results and data transformation. Based on the complexity of the type of measures involved in phase-amplitude coupling analysis, we believe that this degree of risk of bias is acceptable. Specific evaluations for each study were reported in Appendix 1.

Coupling phase

We first assessed the association between the preferred phase of SO-SP coupling and memory retention following sleep. 23 studies (k = 90) were included in the Bayesian hierarchical model. We transformed the circular-linear correlation to standardized coefficients (see Methods: Standardized circular-linear correlation coefficient). Forest and regression plots for overall and moderation models are reported in Figure 4. In addition, the results of hypothesis tests for the overall and moderator models are reported in Table 1 using the Bayes factor and posterior probability. Consistent with the funnel plot (Appendix 4—figure 1A), neither Egger’s regression (p = 0.59) nor rank correlation test (p = 0.52) found the existence of publication bias.

Figure 4

Download asset Open asset

Forest and regression plots for the association between SO-SP coupling phase and memory retention.

(A) Overall model forest plot at study-level. Dashed lines indicate the 95% credible interval (CrI) of the pooled effect size. The black point and error bar for each study show the adjusted estimation of effect size and 95% CrI combining data and prior information. The gray dots under each distribution show raw effect sizes of each study. Effect size-level plots can be found in Supplementary file 3. (B) Meta regression plot with age as moderator. Blue lines represent 200 overplotted spaghetti fit lines to visualize predictions. (C) Moderator-level forest plot. Each box represents a type of moderator. Mixed effect sizes with mixed conditions from different factor levels listed above. *Weight* Stacked weight of each moderation model in the paired model performance comparison between the moderator and overall (intercept-only) model. The stacked weight of the overall model in each pair of comparisons can be calculated as 1 - weight of the moderation model.

Figure 4—source data 1 Subtable of effect size-level metadata included in the coupling phase–memory analysis.: https://cdn.elifesciences.org/articles/101992/elife-101992-fig4-data1-v1.zip
Download elife-101992-fig4-data1-v1.zip

Table 1

Result of directional hypothesis tests for each pair of factor levels (conditions) in overall and each moderation model of the coupling phase-memory association.

Moderator	Overall	Memory Task			Age	Spindle	PSG Channel			Stage	Bout
Condition	H₁	Verbal	Emotional	Spatial	Younger	Fast	Frontal	Frontal	Central	N2	Night
Control	H₀	Motor	Motor	Motor	Older	Slow	Central	Posterior	Posterior	SWS	Nap
BF₁₀	58.35	1.30	1.09	0.07	160.94	11.39	13.58	6.13	0.86	1.74	2.23
Probability	0.98	0.57	0.52	0.06	0.99	0.92	0.93	0.86	0.46	0.63	0.69

Condition conditions hypothesized to be associated with stronger phase-memory association than other factor levels; Control Variables hypothesized to be associated with weaker phase-memory association; BF₁₀ Bayes factor in favor of H₁ over H₀; N2 nREM2 stage; SWS slow-wave sleep.

Overall model

The analysis of a random effect model on the overall phase-memory association revealed that, without considering moderating factors, very strong and consistent evidence supports a small-sized association between the preferred coupling phase and memory consolidation across studies, r_z,pooled = 0.07 [0.01, 0.13], BF₁₀ = 58.35, probability = 0.98. This implies that the likelihood of H₁ is over 58 times greater than H₀, covering approximately 98% of posterior samples.

Multilevel model analysis indicates a heterogeneity similar to typical correlational meta-analyses (Van Erp et al., 2017) (M_g = 0.13), including a between-study heterogeneity of g = 0.07 [0.00, 0.16], as well as a within-study heterogeneity of g = 0.04 [0.00, 0.11]. Focal analysis showed no difference from the original overall model, with an effect size r_z,pooled = 0.07 [−0.06, 0.19]. Sensitivity analysis regarding prior robustness also revealed similar results (see Appendix 3—table 1).

Moderator and sensitivity models

Sufficient evidence shows that moderators including memory tasks, age, spindle types, and PSG channels provide additional predictive power with strong favor for each hypothesis (all contain BF₁₀≤ 0.1 or ≥ 10). Figure 4C indicates that most mixed conditions have a considerably higher uncertainty compared to other normal conditions.

Memory task

Contrary to the assumption, there is no evidence to support a difference of phase-memory association between motor memory (k = 36) and verbal memory retention (k = 35, r_z = Δ0.01, BF₁₀ = 1.30 , probability = 0.57), as well as between motor and emotional memory retention (k = 12, r_z = Δ0.01, BF₁₀ = 1.09, probability = 0.52). However, strong evidence supports that spatial tasks have a considerably lower phase-memory association compared to motor tasks (k = 7, r_z = Δ0.19, BF₁₀ = 0.07, probability = 0.06). The task model has a weight of 0.29, which indicates that the moderator effect of memory task is relatively weak and the task model performed worse than the overall model.

Participant age

As shown in Figure 4B, extremely strong evidence supports that with the increase of age, the slope of correlation exhibits a strongly decreasing trend ( $r_{z β} = Δ - 0.006 [- 0.010, - 0.001]$ [-0.010, -0.001], $B F_{10} = 160.94$ , probability = 0.99). Each ten-year increase in age is associated with a decrease in effect size of 0.06, which is consistent with our hypothesis that precise SO-SP coupling becomes less predictive of memory retention with the increase of age. The moderator effect of age becomes less pronounced during development after excluding the older adult data that represent aging effects, $r_{z β} = Δ - 0.005 [- 0.013, 0.004]$ [-0.013, 0.004], $B F_{10} = 5.51$ , showing a moderate effect. Accounting for the factor of age group significantly increased the predictive power compared to the intercept-only model, given a stacked weight of 1.00.

Spindle frequency

Moderation model with SP frequency range highlights a stronger association between memory retention and SO-fast SP coupling phase $(k = 43)$ rather than slow SPs ( $k = 28$ , $r_{z} = Δ - 0.07$ , $B F_{10} = 11.39$ , probability = 0.92), consistent with the hypothesis based on studies comparing the role of different SP types on the memory retention (Bastian et al., 2022; Mölle and Born, 2011b). The SP model performed as well as the overall model, weight = 0.48.

PSG channel

The strongest pooled correlation was observed in the frontal electrode clusters $(k = 29)$ , compared to the central ( $k = 36$ , $r_{z} = Δ - 0.08$ , $B F_{10} = 13.58$ , probability = 0.93), and posterior channels ( $k = 22$ , $r_{z} = Δ - 0.07$ , $B F_{10} = 6.13$ , probability = 0.86), which represents a moderate-to-strong moderation effect. The frontal area has the largest phase-memory association, $r_{z} = 0.12 [0.03, 0.21]$ [0.03, 0.21]. Effect sizes in posterior areas do not differ from central areas ( $r_{z} = Δ - 0.01$ , $B F_{10} = 0.86$ , probability = 0.46).

Sleep stage and bout

The sleep stage appeared to be a weak predictor with a weight of 0.00, in which the nREM2 stage $(k = 18)$ could not predict a higher phase-memory association than SWS ( $k = 30$ , $r_{z} = Δ - 0.03$ , $B F_{10} = 1.74$ , probability = 0.63). The sleep bout also revealed no difference in phase-memory association between overnight sleep $(k = 73)$ and nap condition ( $k = 17$ , $r_{z} = Δ - 0.03$ , $B F_{10} = 2.23$ , probability = 0.69), with no predictive role (weight = 0.00) compared to the overall model.

Spatiotemporal analysis of coupling phase

By aggregating individual-level data across studies, we found that phase of SO-fast SP coupling has a significant quadratic association with memory retention in frontal regions (see Figure 5A), $r = 0.26$ , $r_{z} = 0.20$ , $p < 0.01$ . However, this association is not significant in either central ( $r = 0.11$ , $r_{z} = 0.05$ , $p = 0.22$ ) or posterior regions ( $r = 0.08$ , $r_{z} = 0.01$ , $p = 0.47$ ).

Figure 5

Download asset Open asset

Preferred slow oscillation-fast spindle coupling phase and its association with memory retention.

(A) Quadratic regression of the phase-memory association under different regions of PSG channels aggregated from studies included in the meta-analysis. 0 peak of SO upstate; ±π trough of SO downstate; r circular-linear correlation coefficient; *r_z* standardized circular-linear correlation coefficient. Bars represent the mean memory retention scores per π/4 radian (45°). The dashed vertical line represents the mean preferred phase across studies. The colored quadratic fit line represents the direction of the relationship. The direction of their relationship gradually flips as the PSG channel moves from the front to the posterior area. Only under the frontal and central channels do fit lines display a quadratic relationship, with a peak of memory score near the up-state peak of SOs. In posterior channels, in contrast, the relationship is convex, although it is not significant. Non-significant quadratic regressions between SO-slow SP coupling and memory are reported in Appendix 6. (B) Posterior distributions of mean preferred phases from the Bayesian circular mixed-effect model. The circular posterior distribution is shown in the top-right corner, and the area between two black lines is projected on a linear scale in the main graph. The vertical line reflects the up-state peak of SOs. Points and error bars denote the mean and 95% credible intervals of phases detected from each channel cluster. Phase values are reported as radians. (C) Circular plot of the preferred coupling phase. From top to bottom, frontal, central, and posterior. The direction of each colored dot represents the preferred coupling phase of each subject recorded from PSG channels in each cluster. The direction of the mean resultant vector indicates the mean preferred coupling phase across subjects, the width indicates the 95% credible interval of the mean coupling phase, the length from 0 (center) to 1 (circumference) indicates the consistency of coupling phase across subjects.

Another important finding is a considerable shift of the preferred coupling phase from frontal to posterior regions, similar to Kurz et al., 2021. After taking into account the repeated measurement, the frontal area has a preferred phase around 0.13 rad [0.00, 0.27], which is the closest to the peak of SO (0 rad). The coupling consistently happened earlier when moving towards further dorsally, reflected by the phase in central (−0.27 rad [−0.37, −0.18]) and posterior area (−0.41 rad [−0.54,−0.27]).

Extremely strong evidence supports a considerable difference of the preferred coupling phase between frontal and central areas, $Δ = - 0.40$ rad, $B F_{10} = + \infty$ , probability = 1.00; between frontal and posterior areas, $Δ = - 0.54$ rad, $B F_{10} = + \infty$ (An ‘infinite’ BF₁₀ value indicates that all posterior samples are overwhelmingly compatible with H₁, given the data and priors. It reflects a reporting convention where BF₁₀ becomes unbounded as the likelihood under H₀ approaches zero.), probability = 1.00; and between central and posterior areas, $Δ = - 0.14$ rad, $B F_{10} = 34.4$ , probability = 0.97. Differences can be observed from the shift of the posterior distribution in Figure 5B and the location of data clusters in Figure 5C. Moreover, the consistency of coupling phase across subjects also decreases from frontal ( $z = 0.69$ , $p < 0.05$ , Rayleigh test) to posterior areas ( $z = 0.59$ , $p < 0.05$ , Rayleigh test). The differences in coupling phase can explain part of the discrepancy of the timing of coupling occurrence among previous studies (Bastian et al., 2022; Hahn et al., 2020; Muehlroth et al., 2019; Joechner et al., 2023).

In summary, we observed that SO-fast SP coupling in the frontal region occurred at the latest phase observed across all regions, characterized by the highest precision and strength of phase-locking. We also found the strongest phase-memory association in frontal regions following a typical quadratic relationship.

Spindle amplitude

We next assessed the association between the amplitude/power of SP and memory retention during the learning night. 18 original studies $(k = 78)$ were included in the Bayesian hierarchical model. The mean amplitude of fast SPs (in μV) is 33.99 [26.44,41.55], while the mean amplitude of slow SPs is 39.41 [33.61, 45.22]. Forest and regression plots for the overall and moderation models are reported in Figure 6. In addition, the results of hypothesis tests for the overall and moderator models are reported in Table 2. The publication bias of the amplitude-memory association studies is unclear but potentially provides evidence of asymmetry. Egger’s regression test $(p = 0.02)$ indicated potential publication bias, whereas the rank correlation test $(p = 0.11)$ and funnel plot (Appendix 4—figure 1C) did not suggest significant bias.

Figure 6

Download asset Open asset

Forest and regression plots for the association between SP amplitude and memory retention.

(A) Overall model forest plot at study-level. The solid vertical line represents the mean Pearson correlation coefficient under the null hypothesis. Dashed lines indicate the 95% credible interval (CrI) of the pooled effect size. The black point and error bar for each study show the adjusted estimation of effect size and 95% CrI combining data and prior information. The gray dots under each distribution show raw effect sizes of each study. Effect size-level plots can be found in Supplementary file 3. (B) Meta regression plot with age as moderator. Blue lines represent 200 overplotted spaghetti fit lines to visualize predictions. (C) Moderator-level forest plot. Each box represents a type of moderator. Mixed effect sizes with mixed conditions from different factor levels listed above. *Weight* Stacked weight of each moderation model in the paired model performance comparison between the moderator and overall (intercept-only) model. The stacked weight of the overall model in each pair of comparisons can be calculated as 1 weight of the moderation model.

Figure 6—source data 1 Subtable of effect size-level metadata included in the spindle amplitude memory analysis.: https://cdn.elifesciences.org/articles/101992/elife-101992-fig6-data1-v1.csv
Download elife-101992-fig6-data1-v1.csv

Table 2

Result of directional hypothesis tests for each pair of factor levels (conditions) in overall and each moderation model of the SP amplitude-memory association.

Moderator	Overall	Memory Task			Age	Spindle	PSG Channel			Stage	Bout
Condition	H₁	Verbal	Emotional	Spatial	Younger	Fast	Frontal	Frontal	Central	N2	Night
Control	H₀	Motor	Motor	Motor	Older	Slow	Central	Posterior	Posterior	SWS	Nap
BF₁₀	8.28	22.19	1.50	9.30	3.70	12.21	1.88	24.67	13.65	5.37	2.54
Probability	0.89	0.96	0.60	0.90	0.79	0.92	0.65	0.96	0.93	0.84	0.72

Condition conditions hypothesized to be associated with stronger amplitude-memory association than other factor levels; Control Variables hypothesized to be associated with weaker amplitude-memory association; BF₁₀ Bayes factor in favor of H₁ over H₀; N2 nREM2 stage; SWS slow-wave sleep.