Sleep spindle maturity promotes slow oscillation-spindle coupling across child and adolescent development

  1. Ann-Kathrin Joechner  Is a corresponding author
  2. Michael A Hahn
  3. Georg Gruber
  4. Kerstin Hoedlmoser
  5. Markus Werkle-Bergner  Is a corresponding author
  1. Center for Lifespan Psychology, Max Planck Institute for Human Development, Germany
  2. Department of Psychology, Laboratory for Sleep, Cognition and Consciousness Research, University of Salzburg, Austria
  3. Centre for Cognitive Neuroscience Salzburg (CCNS), University of Salzburg, Austria
  4. Hertie-Institute for Clinical Brain Research, University Medical Center Tuebingen, Germany
  5. Department of Psychiatry and Psychotherapy, Medical University of Vienna, Austria
  6. The Siesta Group, Austria

Abstract

The synchronization of canonical fast sleep spindle activity (12.5–16 Hz, adult-like) precisely during the slow oscillation (0.5–1 Hz) up peak is considered an essential feature of adult non-rapid eye movement sleep. However, there is little knowledge on how this well-known coalescence between slow oscillations and sleep spindles develops. Leveraging individualized detection of single events, we first provide a detailed cross-sectional characterization of age-specific patterns of slow and fast sleep spindles, slow oscillations, and their coupling in children and adolescents aged 5–6, 8–11, and 14–18 years, and an adult sample of 20- to 26-year-olds. Critically, based on this, we then investigated how spindle and slow oscillation maturity substantiate age-related differences in their precise orchestration. While the predominant type of fast spindles was development-specific in that it was still nested in a frequency range below the canonical fast spindle range for the majority of children, the well-known slow oscillation-spindle coupling pattern was evident for sleep spindles in the adult-like canonical fast spindle range in all four age groups—but notably less precise in children. To corroborate these findings, we linked personalized measures of fast spindle maturity, which indicate the similarity between the prevailing development-specific and adult-like canonical fast spindles, and slow oscillation maturity, which reflects the extent to which slow oscillations show frontal dominance, with individual slow oscillation-spindle coupling patterns. Importantly, we found that fast spindle maturity was uniquely associated with enhanced slow oscillation-spindle coupling strength and temporal precision across the four age groups. Taken together, our results suggest that the increasing ability to generate adult-like canonical fast sleep spindles actuates precise slow oscillation-spindle coupling patterns from childhood through adolescence and into young adulthood.

Editor's evaluation

This is an important analysis of sleep datasets across different age groups that contributes to our understanding of sleep spindle and slow oscillation dynamics during development. The work is expected to be of interest to interdisciplinary fields including development and sleep. The analyses are solid and adequately complex to capture the changes in sleep spindle to slow oscillation coupling among the age groups.

https://doi.org/10.7554/eLife.83565.sa0

eLife digest

Cells in the brain are wired together like an electric circuit that can relay information from one area of the brain to the next. Even when sleeping, the human brain continues to send signals to process information it has encountered during the day. This results in two patterns of electrical activity that define the sleeping brain: slowly repeating waves (or slow oscillations) and rapid bursts of activity known as sleep spindles.

Although slow oscillations and sleep spindles are generated in different regions of the brain, they often happen at the same time. This syncing of activity is thought to help different parts of the brain to communicate with each other. Such communication is essential for new memories to become stable and last a long time.

In children, slow oscillations and sleep spindles appear together less frequently, suggesting that these co-occurring patterns of electrical activity develop as humans grow into adults. Here, Joechner et al. set out to understand what drives slow oscillations and sleep spindles to start happening at the same time.

The team used a technique called electroencephalography (or EEG for short) to study the brain activity of children, teenagers and adults as they slept. This revealed that slow oscillations and sleep spindles occur together less often in children compared to teenagers and adults. Moreover, the slow oscillations and sleep spindles observed in the children had very different physical characteristics to those observed in adults. Further analyses showed that the more similar the children’s sleep spindles were to adult spindles, the more consistently they appeared at the same time as the slow oscillations.

The findings of Joechner et al. suggest that as children grow up, their sleep spindles become more adult-like, causing the spindles to happen at the same time as slow oscillations more consistently. This indicates that brain circuits that generate sleep spindles may play an essential role in developing successful communication networks in the human brain. In the future, this work may ultimately provide new insights into how age-related changes to the brain contribute to cognitive development, and suggests sleep as a potential intervention target for neurodevelopmental disorders.

Introduction

The grouping of sleep spindles (9–16 Hz; Cox et al., 2017) into sequences of increased and decreased activity by the sleep slow oscillation (SO, 0.5–1 Hz; Steriade, 2006) during non-rapid eye movement sleep (NREM) has been recognized as an intrinsic property of the healthy, mature mammalian corticothalamic system for decades (Contreras et al., 1996; Contreras and Steriade, 1995; Mulle et al., 1986; Staresina et al., 2015; Steriade et al., 1993). The joint depolarization of large groups of cortical neurons during the SO up state impinges on neurons of the reticular thalamic nucleus, there, creating conditions that facilitate thalamic spindle generation (Steriade, 1999). Sleep spindles then propagate to the cortex via thalamocortical projections, where they promote synaptic plasticity through changes in calcium activity (Niethard et al., 2018; Rosanova and Ulrich, 2005). Yet, little is known about how this precise coalescence develops across childhood and adolescence.

Far from being an epiphenomenon, accumulating evidence suggests that the synchronization of canonical fast sleep spindles (i.e., spindles defined in young adults with a frequency of ≈ 12.5–16 Hz and a centro-parietal predominance; Cox et al., 2017) precisely during the up state of SOs provides an essential mechanism for neural communication, for example, supporting systems memory consolidation during sleep (Hahn et al., 2020; Helfrich et al., 2018; Latchoumane et al., 2017; Mölle et al., 2002; Muehlroth et al., 2019). Importantly, canonical fast spindles in turn assort hippocampal ripples (Clemens et al., 2007; Helfrich et al., 2019; Siapas and Wilson, 1998; Staresina et al., 2015), that code for wake experiences and are considered a reliable marker of hippocampal memory consolidation (Buzsáki, 2015; Maingret et al., 2016; Sirota et al., 2003). Moreover, canonical fast sleep spindles themselves are associated with facilitated hippocampal-neocortical connectivity (Andrade et al., 2011; Cowan et al., 2020). In addition to canonical fast sleep spindles, there is substantial evidence for a canonical slow sleep spindle type (i.e., spindles defined in young adults with a frequency of ≈ 9–12.5 Hz and a frontal predominance; Cox et al., 2017) in the human surface electroencephalogram (EEG; De Gennaro and Ferrara, 2003; Fernandez and Lüthi, 2020). That said, previous findings hint at a differential SO-slow spindle coupling pattern and their function is still elusive (Klinzing et al., 2016; Mölle et al., 2011; Muehlroth et al., 2019; Rasch and Born, 2013). Taken together, the complex wave sequence of SO up state and canonical fast sleep spindles, together with hippocampal activity, is considered to provide the scaffold for the precisely timed reactivation of initially fragile hippocampal memory representations and their strengthening in neocortical networks (Diekelmann and Born, 2010; Helfrich et al., 2019; Maingret et al., 2016; Staresina et al., 2015). However, the precise coupling of sleep spindle activity to SOs, described above, does not seem to be fully present and functional from early childhood on. Recent evidence in rodents and humans indicates that the temporal co-ordination of SOs and spindles improves across childhood and adolescence (García-Pérez et al., 2022; Hahn et al., 2020; Joechner et al., 2021).

Likewise, the individual neural rhythms that define the coupling undergo substantial changes during child and adolescent development. Across maturation, sleep spindles increase in occurrence and their average frequency (Purcell et al., 2017; Zhang et al., 2021). Consistent with this, canonical slow sleep spindles were reported to mature and dominate during early childhood. In contrast, canonical fast spindles are rarely detected in young children and become increasingly present and clearly dissociable only around puberty (D’Atri et al., 2018; Goldstone et al., 2019; Hoedlmoser et al., 2014; Shinomiya et al., 1999). However, amongst others, the application of individually adjusted frequency bands revealed that already young children express functional fast spindles in centro-parietal sites, whereby these manifest in a development-specific fashion (D’Atri et al., 2018; Friedrich et al., 2019; Joechner et al., 2021; Zhang et al., 2021). Individualized rhythm detection methods provide an effective approach to capture true, dominant oscillatory rhythms despite substantial inter-individual variability, which presents a particular methodological challenge in developmental and age-comparative research (Cox et al., 2017; Muehlroth and Werkle-Bergner, 2020). While canonical fast spindles become more pronounced across development, an opposite trend can be observed for SOs (Buchmann et al., 2011; Kurth et al., 2010a). Slow neuronal activity is initially maximally expressed and originates over posterior areas, developing towards a mature anterior predominance (Kurth et al., 2010b; Timofeev et al., 2020). In summary, paralleling developmental changes in SO-spindle coupling, fast sleep spindles and SOs separately are manifested differentially across development. Nevertheless, it is still unclear how developments in sleep spindles and SOs interact to promote precise, adult-like temporal synchronization of sleep spindles during SOs across childhood and adolescence.

Therefore, we aimed to (i) characterize the modulation of sleep spindles during SOs across different ages and (ii) investigate how sleep spindle and SO maturity relate to the manifestation of SO-spindle coupling across different ages. Specifically, based on previous analyses (Joechner et al., 2021), we reasoned that the development of fast sleep spindles might be associated with the maturation of SO-spindle coupling. For this, we re-analyzed previously published nocturnal EEG data from a cross-sectional sample of 24 5- to 6-year-old children (13 female, Mage = 5 years, 10.71 months, SDage = 7.28 months; Joechner et al., 2021) and a longitudinal cohort of 33 children tested at 8–11 years of age (T1; Hoedlmoser et al., 2014) and again at 14–18 years of age (T2; 23 female, MageT1 = 9 years, 11.70 months, SDageT1 = 8.35 months; MageT2 = 16 years, 4.91 months, SDageT2 = 9.06 months; Hahn et al., 2019; Hahn et al., 2020). Further, we examined a cross-sectional sample of 18 adults aged 20–26 years (15 female, Mage = 21 years, 8.78 months, SDage = 20.04 months) as a reference sample for adult-like patterns. All cohorts underwent two nights of ambulatory polysomnography (PSG). Given well-known first night effects in children (Scholle et al., 2003), sleep spindles, SOs, and their coupling were only analyzed during the second night here.

Results

Dominant fast centro-parietal sleep spindles become more prevalent and increasingly resemble canonical fast sleep spindles with older age

In the first step, we aimed at identifying age-specific patterns of sleep spindles. To obtain evidence for two distinct dominant oscillatory spindle rhythms within all four age groups (5- to 6-, 8- to 11-, 14- to 18-, 20- to 26-year-olds), we firstly determined individual spindle peak frequencies (between 9–16 Hz) of background-corrected power spectra during non-rapid eye movement sleep (NREM, N2 and N3) at averaged frontal (F3, F4) and centro-parietal (C3, C4, Cz, Pz) electrodes, where slow and fast sleep spindles typically dominate, respectively (Cox et al., 2017; for background-corrected power spectra, see Figure 1A; see Supplementary file 1a and b for statistical comparisons of peak frequencies). Based on the individual frontal and centro-parietal spindle peak frequencies, we then detected sleep spindle events in frontal and centro-parietal sites which represent the person- and age-specific dominant spindle rhythms per individual (see Figure 1—figure supplement 1 for examples of averaged EEG signals time-locked to the occurrence of these dominant sleep spindles). We then compared features of these individually identified sleep spindles (i.e., frequency, density, amplitude; Figure 1B–D) between age groups (5- to 6-, 8- to 11-, 14- to 18-, 20- to 26-year-olds) and topographies (frontal, centro-parietal) using linear mixed-effects models (LMM):

spindlefeature1+agegrouptopography+(1|ID)
Figure 1 with 2 supplements see all
Sleep spindle features across the different age groups.

(A) Background-corrected power spectra in averaged frontal (left) and averaged centro-parietal (right) electrodes for every participant at every test time point (y-axis). Power is coded by color for frequencies between 5–20 Hz (x-axis). The individual peak frequency in the sleep spindle frequency range (x-axis, 9–16 Hz) is reflected by brighter colors for every participant at every age (y-axis). Data are ordered by age from bottom to top (y-axis). The white dotted line at 12.5 Hz illustrates the frequency border for canonical fast sleep spindles (i.e., spindles defined in young adults with a frequency ≈ 12.5–16 Hz and a centro-parietal predominance; Cox et al., 2017). Note: Participants at an older age (top) showed higher peak frequencies. Further, a linear mixed-effects model revealed that peak frequencies in centro-parietal electrodes were significantly faster compared to frontal ones averaged across all age groups (Supplementary file 1b). However, for most of the young children (bottom part of the plot) the peak frequencies and the frequencies of the derived dominant fast spindles (B) were slower than the canonical fast sleep spindle band (i.e., development-specific). (B–D) Results from the comparison of adult-like fast (12.5–16 Hz) and individually identified slow frontal and development-specific fast centro-parietal sleep spindle (B) frequency, (C) density, and (D) amplitude for all four age groups. (B) Horizontal lines at 12.5 Hz illustrate the frequency border for canonical fast sleep spindles (i.e., spindles defined in young adults with a frequency ≈ 12.5–16 Hz and a centro-parietal predominance; Cox et al., 2017). Values are individual raw scores. Diamonds reflect estimated marginal means of the respective linear mixed-effects model. Statistical results can be found in Supplementary files 1a–n.

The results from the LMMs were further specified using pairwise comparisons for which p-values were corrected using the Bonferroni method (padj; Bland and Altman, 1995).

For all features of individually identified, dominant sleep spindles, all main effects and the interaction between age group and sleep spindle topography reached significance (Ffrequency(3, 134.73) = 4.04, pfrequency = 0.009; Fdensity(3, 136.76) = 11.73, pdensity < 0.001; Famplitude(3, 140.21) = 5.45, pamplitude = 0.001; Supplementary file 1c, e, and g). For sleep spindle frequency, follow-up pairwise comparisons revealed that sleep spindles showed a higher frequency at centro-parietal derivations as compared to frontal electrodes within every age group (all t ≤ –5.38, all padj < 0.001). Further, frontal spindle frequency was lowest for the 5- to 6- year-olds as compared to all older age groups (all t ≤ –3.69, all padj ≤ 0.010) and lower for the 8- to 11-year-olds as compared to the older age groups (all t ≤ –3.61, all padj ≤ 0.013), while there was no difference between the 14- to 18- and the 20- to 26-year-olds (t(112.39) = 0.09, padj = 1.000). For centro-parietal sleep spindles there was no frequency difference between the 5- to 6- and the 8- to 11-year-olds (t(112.39) = –3.03, padj = 0.084) and the 14- to 18- and the 20- to 26-year-olds (t(112.39) = –2. 03, padj = 1.000; for all pairwise comparison results, see Supplementary file 1d).

Hence, across all age groups, individually identified frontal spindles revealed lower frequencies than centro-parietal sleep spindles, indicating the presence of two distinguishably fast spindle types (note, the same applied to the peak frequencies, see Supplementary file 1a and b). Thus, we will henceforth be referring to slow, that is, frontal, and fast, that is, centro-parietal, spindles, respectively. Crucially, despite being faster, the frequency of the fast centro-parietal spindles was specific to the age of the participants in a way that at younger ages, these dominant fast sleep spindles were not yet in the range of canonical fast spindles (i.e., as on average observed in adults with a frequency of ≈ 12.5–16 Hz; Cox et al., 2017; see Figure 1A–B). For the vast majority of 14- to 18- and 20- to 26-year-olds, individually identified, dominant centro-parietal sleep spindles indeed matched the canonical fast sleep spindle frequency range (≈ 12.5–16 Hz, see Figure 1A–B). For the majority of children though, dominant fast centro-parietal spindles were nested in the canonical slow spindle range (i.e., as on average observed in adults with a frequency of ≈ 9–12.5 Hz; Cox et al., 2017) and thus manifested in a development-specific fashion. Hence, the term ‘development-specific’ will be employed to refer to individually determined, dominant fast centro-parietal sleep spindles across all age groups in the following. In contrast, since we also found dominant fast centro-parietal sleep spindles between 12.5 and 16 Hz in our adult sample, we will denote sleep spindle activity in this canonical fast spindle range and events detected exclusively within this canonical frequency range in centro-parietal sites as ‘adult-like’ in our sample across all age groups.

For density (Figure 1C; Supplementary file 1f), pairwise comparisons indicated higher density for slow frontal compared to development-specific fast centro-parietal spindles for the 5- to 6- (t(136.76) = 5.69, padj < 0.001) and the 8- to 11-year-olds (t(136.76) = 4.62, padj < 0.001), while the difference was not significant for 14- to 18- and the 20- to 26-year-olds (t14-18(136.76) = 0.83, padj 14-18 = 1.000; t20-26(136.76) = –2.07, padj 20-26 = 1.000). For spindle amplitude (Figure 1D, Supplementary file 1h), amplitudes were significantly higher for slow frontal as compared to development-specific fast centro-parietal spindles within all age groups (all t ≥ 4.74, all padj ≤ 0.001). The interaction was mainly driven by a significantly higher slow frontal amplitude for the 8- to 11-year-olds compared to all other age groups (all –3.68 ≥ t ≥ 7.97, all padj ≤ 0.009), a higher slow frontal amplitude for the 5- to 6-year-olds compared to the 20- to 26-year-olds (t(137) = 4.32, padj < 0.001), a lower development-specific fast centro-parietal amplitude for the 14- to 18- compared to the 8- to 11-year-olds (t(140) = 5.24, padj < 0.001), and a lower amplitude of the 20- to 26-year-olds compared to the 5- to 6- and 8- to 11-year-olds (t ≥ 3.34, padj ≤ 0.030).

To sum up, our data indicate the pronounced existence of slow frontal and development-specific fast centro-parietal spindles at all ages under study. In particular, development-specific fast centro-parietal spindles were more numerous and faster at older age, especially compared to slow frontal sleep spindles. Given evidence for the specific role of canonical fast spindles for memory (Rasch and Born, 2013), henceforth, we focus on fast spindles detected at centro-parietal electrodes (however, corresponding analyses were also conducted for slow frontal spindles and can be found in Supplementary file 1).

After having characterized prevailing, development-specific fast centro-parietal spindles across all our age groups, we were interested in how they differed from canonical fast spindles, that is, those commonly, and also here, found in young adults with a frequency of ≈ 12.5–16 Hz (see Figure 1 for the present adult sample, Cox et al., 2017; Ujma et al., 2015). Therefore, we additionally extracted fast spindles in centro-parietal electrodes by applying fixed frequency criteria between 12.5 and 16 Hz (henceforth, adult-like fast sleep spindles; see Figure 1—figure supplement 2 for an example raw EEG trace with development-specific and adult-like fast spindles and Figure 1—figure supplement 1 for examples of averaged EEG signals time-locked to development-specific and adult-like fast sleep spindles). Such adult-like fast spindles were shown before to be coupled to SOs in pre-school children, despite lacking evidence for their strong presence (Joechner et al., 2021; see also Figure 1A, Figure 3, and Figure 1—figure supplement 1). We then compared characteristics (i.e., frequency, density, amplitude; Figure 1B–D) of development-specific fast centro-parietal sleep spindles with adult-like fast centro-parietal spindles using LMMs and follow-up pairwise comparisons across the four age groups. In addition to ‘age group’, the factor ‘spindle type’ (development-specific, adult-like) was entered as a fixed factor:

spindlefeature1+agegroupspindletype+(1|ID)

Results indicated that all main effects and the interaction between the factors ‘age group’ and ‘spindle type’ were significant for frequency, density, and amplitude (for a summary of all analyses and all post-hoc comparisons see Supplementary file 1i–n). Post-hoc pairwise comparisons revealed that the frequency of adult-like fast sleep spindles was consistently higher as compared to the development-specific fast centro-parietal spindles within all age groups, except the 20- to 26-year-olds (t5-6(135.85) = 18.70, padj 5-6 < 0.001; t8-11(135.85) = 12.48, padj 8-11 < 0.001; t14-18(135.85) = 5.37, padj 14-18 < 0.001; t20-26(135.85) = 2.73, padj 20-26 = 0.199). In line with a generally higher frequency, in 5- to 6- and 8- to 11-year-olds, adult-like fast spindles had a lower amplitude compared to development-specific fast centro-parietal spindles (t5-6(142.56) = –4.34, padj 5-6 < 0.001; t8-11(142.56) = –4.47, padj 8-11 < 0.001). Crucially, this effect was non-significant in the two oldest age groups (t14-18(142.56) = –0.99, padj 14-18 = 1.000; t20-26 (142.56) = 0.14, padj 20-26 = 1.000). Similarly, whereas the density of development-specific fast sleep spindles was higher compared to adult-like fast spindle density in the 5- to 6- and the 8- to 11-year-olds (t5-6(138.60) = –7.66, padj 5-6 < 0.001; t8-11(138.60) = –9.69, padj 8-11 < 0.001), there was no difference in density for the 14- to 18- and 20- to 26-year-olds (t14-18(138.60) = –1.50, padj 14-18 = 1.000; t20-26(138.60) = –0.13, padj 20-26 = 1.000). Further, density of adult-like fast sleep spindles was higher for the two oldest age groups compared to the 8- to 11- and 5- to 6-year-olds and higher for the 8- to 11- as compared to the 5- to 6-year-olds (all t ≤ –4.15, all padj ≤ 0.002). Density for the development-specific fast centro-parietal spindles was only lower for the 5- to 6-year-olds as compared to all older age groups (t8-11(113.55) = –5.35, padj 8-11< 0.001; t14-18(113.55) = –5.32, padj 14-18< 0.001; t20-26(113.55) = –4.15, padj 20-26 = 0.002), while there was no difference between the 8- to 11-, the 14- to 18, and the 20- to 26-year-olds (0 < t ≤ 0.48; all padj = 1.000; see Supplementary file 1j, l, and n for all comparisons).

Crucially, this indicates that despite the absence of a prominent peak in the power spectrum (Figure 1A), already young children express adult-like fast spindles which occur increasingly often with older age. Despite a higher frequency of the adult-like fast sleep spindles in the child and adolescent age groups, with older age, the differences in amplitude and density between development-specific and adult-like fast centro-parietal sleep spindles decreased and were no more evident in the adolescent and adult age groups. Hence, fast centro-parietal spindles become more numerous and adult-like in their frequency and amplitude characteristics in the course of maturation.

Slow oscillations dominate anteriorly in older children, adolescents, and young adults but not in young children

Similar to the analyses on sleep spindles, we next compared features that define SOs (i.e., frequency, density, amplitude; Figure 2) between age groups and between SOs detected at frontal (averaged over F3, F4), centro-parietal (averaged over Cz, Pz), and occipital (cross-sectional child sample: Oz, longitudinal and cross-sectional adult samples: averaged over O1, O2) electrodes (topography factor) using LMMs and follow-up pairwise comparisons:

SOfeature1+agegrouptopography+(1|ID)
Slow oscillation features across the different age groups.

Results from the comparison of slow oscillation (A) frequency, (B) density, and (C) amplitude for all four age groups. Values are individual raw scores. Diamonds reflect estimated marginal means of the respective linear mixed-effects model. Note that we observed the strongest age-related differences for the slow oscillation amplitude (C) in different topographic locations. Statistical results can be found in Supplementary file 1o–t.

Occipital locations were added based on observations suggesting a posterior dominance of SOs in young children (Kurth et al., 2010b; Timofeev et al., 2020). Further, considering evidence indicating that surface SOs might be most powerful medially (Murphy et al., 2009), we focused on midline electrodes, while also keeping the overlap of electrodes between our samples as high as possible.

Results revealed that the frequency of SOs differed over topographic locations across age groups (age group*topography interaction: F(6,238.15) = 13.10, p < 0.001; see also Supplementary file 1o). Pairwise comparisons (Supplementary file 1p) indicated that SO frequency differed between recording sites for the 14- to 18- and the 20- to 26-year-olds. For both age groups frontal and centro-parietal frequency was higher compared to occipital frequency (tfrontal-occipital 14-18(238.15) = 10.48, padj frontal-occipital 14-18 < 0.001; tcentro-parietal-occipital 14-18(238.15) = 8.09, padj centro-parietal-occipital 14-18 < 0.001; tfrontal-occipital 20-26(238.15) = 9.62, padj frontal-occipital 20-26 < 0.001; tcentro-parietal-occipital 20-26(238.15) = 7.26, padj centro-parietal-occipital 20-26 < 0.001; Figure 2A). For SO density, summary statistics indicated significant main effects for ‘age group’ (Supplementary file 1q; F(3,96.72) = 64.23, p < 0.001) and ‘topography’ (F(2,240.58) = 4.29, p = 0.015) but no interaction effect. Pairwise comparisons revealed that density, averaged across topographical recording sites, was lower for the 8- to 11- compared to both the 5- to 6- (t(78.34) = 3.30, padj = 0.009) and the 14- to 18-year-olds (t(240.58) = –13.64, padj < 0.001) and lower for the 20- to 26-year-olds compared to the 14- to 18-year-olds (t(78.34) = 4.27, padj < 0.001; see Supplementary file 1r for all pairwise comparisons). Further, density averaged across all age groups was significantly higher for frontal as compared to centro-parietal derivations (t(241) = 2.93, padj = 0.011; Figure 2B). For SO amplitude, inspection of the summary statistics and the pairwise comparisons revealed that all age groups, except the 5- to 6-year-olds, expressed SOs at a higher amplitude at frontal as compared to both centro-parietal and occipital locations (all t ≥ 6.68, all p < 0.001; see Supplementary file 1s and t for all results; Figure 2C).

To summarize, we did not observe a frontal prevalence of SOs in the youngest children. However, this dominance was present in all older age groups. Therefore, similar to our sleep spindle analyses, we concentrate on centro-parietal SOs for all subsequent analyses. Results for spindles and SOs in additional topographical locations can be found in Supplementary file 1 and the figure supplements.

Temporal modulation of adult-like fast sleep spindle power during slow oscillations is present across all ages

We observed robust age-related differences in sleep spindles and SOs across the four age groups. But do they also affect the temporal coupling between these two neural rhythms at a given age? In the first step, we aimed at determining the spectral and temporal characteristics of SO-coupled spindles. Hence, within the four age groups, we compared spectral power (5–20 Hz) over centro-parietal recording locations during trials with and without centro-parietal SOs (±1.2 s around SO down peak and equally long trials without SOs) using cluster-based permutation tests (Maris and Oostenveld, 2007). Within all age groups, power in a broad range including the sleep spindle frequency band (9–16 Hz) was significantly higher during SOs as compared to trials without SOs (all cluster p < 0.001), suggesting temporal clustering of spindles during the SO cycle (Figure 3). On a descriptive level, the strongest power differences during SOs as compared to trials without SOs for all four age groups were located within the frequency range of adult-like fast sleep spindles (12.5–16 Hz) within one second after the down peak, close to the up peak (Figure 3). However, the modulation of power in this adult-like fast spindle frequency range seemed to be stronger and more precise during the SO up peak in our two oldest age groups. Specifically, for the younger children, the increased power in the adult-like fast spindle frequency range was surprising, given the overall lower density and power of adult-like fast sleep spindles in 5- to 6- and also a majority of 8- to 11-year-old children (Figure 1).

Figure 3 with 1 supplement see all
Centro-parietal power differences between trials with and without centro-parietal slow oscillations (in t-score units).

Significant clusters are outlined in black (cluster-based permutation test, cluster α < 0.05, two-sided test). Warmer colors indicate higher power during trials with slow oscillations and colder colors indicate lower power during trials with slow oscillations as compared to trials without slow oscillations. The average centro-parietal slow oscillation for each age group (A–D) is plotted onto the power differences in black to illustrate the relation to slow oscillation phase (scale in mV on the right y-axis of each plot). The sleep spindle frequency range is highlighted by the dashed window. Note that the strongest power increases during slow oscillations were observed in a frequency range reflecting the adult-like fast sleep spindle range (12.5–16 Hz). Results for frontal power and frontal slow oscillations can be found in Figure 3—figure supplement 1.

At older age, development-specific fast sleep spindle occurrence is more strongly modulated by slow oscillations while slow oscillation-adult-like fast spindle coupling becomes temporally more precise

Having identified evidence for the temporally ordered occurrence of sleep spindles during specific SO phases in all our age groups, in the next step, we were interested in the precise temporal co-ordination of spindle events and SOs. Therefore, we created peri-event time histograms (PETH) of development-specific fast spindles during centro-parietal SOs (± 1.2 s around the SO down peak), showing the occurrence of sleep spindles within 100 ms time bins during the SO down peak-centered time window. To identify patterns of increased and decreased spindle occurrence during the SO cycle, we compared the resulting sleep spindle occurrence-percentage distribution per participant with a participant-specific surrogate distribution using cluster-based permutation tests (Maris and Oostenveld, 2007).

As can be inferred from Figure 4A, we observed a shift towards a clear coupling from the 5- to 6- to the 14- to 18-year-olds. Only the 14- to 18-year-olds reliably presented the coupling pattern known from, and also observed here in, adults (see Figure 4A, rightmost plot): A decreased sleep spindle probability during the SO down state and an increased occurrence during the SO up state. Statistically, we found one positive cluster (p = 0.001) for the 5- to 6-year-olds from –700 to –400 ms for the modulation of development-specific fast centro-parietal spindles (SO up peak at ≈ 465 ms). For the 8- to 11-year-olds, we identified a more extended positive cluster from –1200 to –400 (p < 0.001) and one negative cluster from –300 to 400 ms (p < 0.001) for development-specific fast centro-parietal spindles (SO up peak at ≈ 488 ms). For the 14- to 18-year-olds, the analyses indicated two positive clusters. One from 300 to 800 ms (p = 0.002) and another from –1000 to –600 ms (p = 0.014) and two negative clusters from –600 to 200 ms (p < 0.001) and from 1000 to 1200 ms (p = 0.034) for development-specific fast centro-parietal spindles (SO up peak at ≈ 532 ms). For the 20- to 26-year-olds, there was one significant positive cluster (p = 0.005) from 300 to 700 ms and two negative clusters; one from –500 to 200 ms (p < 0.001) and from 900 to 1200 ms (p = 0.004; SO up peak at ≈ 559 ms).

Figure 4 with 4 supplements see all
Peri-event time histograms for (A) development-specific fast centro-parietal spindles and (B) adult-like fast centro-parietal spindles showing the proportion of events occurring within 100 ms bins during centro-parietal slow oscillations.

Error bars represent standard errors. Green asterisks mark increased spindle occurrence (positive cluster, cluster-based permutation test, cluster α < 0.05, two-sided test) and red asterisks mark decreased spindle occurrence (negative cluster, cluster-based permutation test, cluster α < 0.05, two-sided test) compared to a surrogate distribution representing random occurrence (horizontal line). The dashed vertical line indicates the slow oscillation down peak. The average centro-parietal slow oscillation of each age group is shown in black to illustrate the relation to the slow oscillation phase. Results for slow oscillations and spindles in different topographies can be found in Figure 4—figure supplement 1.

Critically, and somewhat unexpectedly for the two younger age groups, the previous analysis of spectral power differences in the sleep spindle frequency range along the SO cycle suggested temporal SO-spindle alignment specifically for events in the adult-like fast spindle range (Joechner et al., 2021; Piantoni et al., 2013a). Hence, we repeated the PETH analyses for adult-like fast sleep spindles. As can be inferred from Figure 4B, we observed a clear coupling of adult-like fast spindle occurrence rates to specific phases of the SO cycle within all age groups— although temporally less precisely during the SO down and up peaks in the younger age groups.

For 5- to 6-year-olds, we identified one positive cluster (p < 0.001) from 100 to 500 ms and two negative clusters (both p < 0.001) from –500 to 0 ms and from 700 to 1200 ms for adult-like fast centro-parietal spindles (SO up peak at ≈ 465 ms). For the 8- to 11-year-olds, we found two positive clusters from –1000 to –600 ms (p < 0.001) and from 200 to 600 ms (p = 0.001) and two negative clusters (both p < 0.001) from –400 to 100 ms and from 800 to 1200 ms (SO up peak at ≈ 488 ms). For the 14- to 18-year-olds, we identified one positive cluster (p < 0.001) from 200 to 800 ms and two negative clusters from –600 to 200 ms (p < 0.001) and from 900 to 1200 ms (p = 0.004) for adult-like fast centro-parietal spindles (SO up peak at ≈ 532 ms). Lastly, for the 20- to 26-year-olds, analyses revealed one positive cluster (p = 0.003) from 200 to 700 ms and two negative clusters. One cluster (p < 0.001) from –500 to 200 ms and another one (p = 0.004) from 900 to 1200 ms (SO up peak at ≈ 559 ms). Hence, while occurrence patterns during SOs appeared different for development-specific and adult-like fast sleep spindles in the child age groups, modulation patterns during SOs were highly comparable for these two fast spindle types in the two older age groups.

To summarize, a clear coupling of spindle occurrence at specific phases of the SO could hardly be detected for development-specific fast centro-parietal sleep spindles in the younger age groups. Importantly, adult-like fast sleep spindles were already modulated by SOs in the younger age groups—even though they only occurred very rarely. Of note, a randomly selected lower absolute number of adult-like fast sleep spindles in the 14- to 18-year-olds did not affect the co-occurrence pattern, suggesting that enhanced modulation of sleep spindles during SOs does not merely depend on the number of sleep spindles (see Figure 4—figure supplement 2). However, independent of the sleep spindle type, increased sleep spindle occurrence was more precisely timed with the SO up peak at older ages. Hence, despite clearly identifiable development-specific fast spindles, pronounced and temporally precise SO-spindle coupling seems to depend on the presence of adult-like fast sleep spindles.

Slow oscillation-spindle coupling is related to sleep spindle maturity

So far, in line with previous observations (Joechner et al., 2021), both analyses on SO-spindle co-occurrence suggested that sleep spindles in the adult-like fast frequency range rather than the more dominant, development-specific fast sleep spindles are coupled to SOs within all age groups. Hence, we reasoned, that the maturation of dominant, development-specific fast sleep spindles towards adult-like fast spindles may explain age-related differences in the strength and temporal precision of SO-spindle coupling patterns.

To capture the ‘maturational stage’ of fast spindles for every participant, we resorted to our observed age-related differences in sleep spindles (Figure 1) and computed the distance between development-specific and adult-like fast sleep spindles within each individual at a given age. Specifically, we calculated difference measures in sleep spindle characteristics (i.e., frequency, density, amplitude) between adult-like and development-specific fast sleep spindles. Note, given the opposing signs of the fast spindle maturity differences, we inverted the frequency difference values. Further, we re-scaled the frequency, density, and amplitude difference scores using Z-transformation across all participants to convert all metrics into a common space. As illustrated in Figure 5A–C, the increasing Z-transformed difference scores (henceforth termed ‘spindle maturity scores’) with older age of the participants capture the fact that the dominant, development-specific fast centro-parietal sleep spindles resemble more closely adult-like fast spindles at older ages.

Measures of sleep spindle and slow oscillation maturity and slow oscillation-spindle coupling strength across all age groups.

(AC) Fast sleep spindle maturity scores for (A) frequency, (B) density, and (C) amplitude. Maturity scores reflect the Z-standardized differences between adult-like and development-specific fast centro-parietal sleep spindles. (D) Slow oscillation maturity scores represent the Z-standardized difference between frontal and centro-parietal amplitudes. (E) First principal component of a principal component analysis on the three fast spindle maturity scores which are shown in (AC). (F) Kullback-Leibler divergence for development-specific fast centro-parietal sleep spindle modulation during centro-parietal slow oscillations, reflecting slow oscillation-spindle modulation strength. For all measures, higher values are linked to older age. Asterisks illustrate the mean.

Following a similar logic, we also created a measure for SO maturity. Based on our observation that age differences were mostly reflected in the emerging frontal dominance of SOs with older age (Figure 2) and on the literature (e.g., Kurth et al., 2010b; Timofeev et al., 2020), we calculated the difference between the amplitude in frontal and centro-parietal regions to reflect the maturity pattern of SOs within a participant. In parallel with the procedure for the fast spindle maturity scores, we Z-standardized the SO difference measure to be in the same metric space as the spindle maturity values for all subsequent analyses. Comparably to the fast spindle maturity scores, participants of older age showed higher Z-standardized SO difference scores (Figure 5D).

To examine how fast sleep spindle maturity would be associated with SO-spindle coupling patterns across different ages, we conducted two analyses. For one, we examined the relation between the pattern of power modulations in the spindle frequency range (9–16 Hz) during the complete SO trial (down peak ± 1.2 s, time-frequency t-maps, Figure 3) with the fast spindle maturity scores for sleep spindle frequency, density, and amplitude using a partial least squares correlation (PLSC; Krishnan et al., 2011) across all participants. This analysis provides pairs of SO-spindle coupling profiles and associated, multivariate patterns of spindle maturity scores (spindle maturity profile). Based on a permutation test, we identified one significant pair of a spindle maturity profile (Figure 6A) and a specific time-frequency SO-spindle coupling pattern (Figure 6B; singular vector pair p < 0.001). All fast spindle maturity measures contributed reliably and positively to this significant, positive correlation (as indicated by the direction of values in the spindle maturity profile and the non-zero crossing confidence intervals in Figure 6A). As can be inferred from Figure 6B, the SO-spindle coupling pattern suggests, that higher fast sleep spindle maturity in all features (Figure 6A) was associated with a more adult-like SO-spindle coupling pattern, reflected in: (i) lower adult-like fast spindle power and higher power in the canonical slow sleep spindle range during the down state and (ii) higher power in the adult-like fast spindle frequency range during the SO up peak. Overall this indicates that a stronger presence of spindles with more adult-like fast spindle characteristics is associated with the well-known pattern of increased adult-like fast sleep spindle activity during the SO up peak, increased canonical slow spindle activity during the down state, and decreased activity of adult-like fast spindles during the down peak.

Figure 6 with 1 supplement see all
Association between fast spindle maturity and centro-parietal slow oscillation-spindle coupling.

(AB) Results from a partial least squares correlation revealed one (A) fast spindle maturity profile significantly associated with (B) a centro-parietal slow oscillation-spindle coupling pattern. (A) Weights of the first singular vector dimension of the fast spindle maturity scores. Error bars represent 95% bootstrap confidence intervals. (B) Weights of the first singular vector dimension of the slow oscillation-spindle coupling pattern by means of bootstrap ratios. Only values > 1.96 and < –1.96 are colored. Warmer colors represent higher power and colder colors reflect lower power. (C) Scatterplot of the association between the modulation strength (Kullback-Leibler divergence) of development-specific fast centro-parietal sleep spindles during centro-parietal SOs and the fast spindle maturity component. The curved line represents the prediction from the generalized linear mixed-effects model for the simple effect of the spindle maturity component. For visualization purposes, the four age groups are indicated by different shapes and colors. Results for frontal slow oscillations can be found in Supplementary file 1u and Figure 6—figure supplement 1.

After having identified that fast spindle maturity was associated with stronger modulation of spindle power during the SO cycle, we further examined whether fast spindle maturity would also be related to the strength of modulation of development-specific fast centro-parietal sleep spindles during SOs (see PETH analyses, Figure 4A). Since all three indicators of fast sleep spindle maturity (frequency, density, amplitude maturity) showed age differences and were related to the pattern of spindle power modulations during SOs (Figure 6A), we used principal component analysis (PCA) to create one latent component capturing the maximal amount of variance across our fast spindle maturity indicators (as for individual sleep spindle maturity scores, higher component values indicate higher fast spindle maturity, Figure 5E). This fast spindle maturity component was then used to examine the relation between fast spindle maturity and development-specific fast spindle modulation strength across all participants at every age. We quantified the strength of modulation of development-specific fast centro-parietal sleep spindles during centro-parietal SOs using the Kullback-Leibler (KL) divergence for which higher values reflect a stronger clustering of sleep spindles around specific phases of SOs (Figure 5F, higher values are associated with older age). We then conducted a log-linked gamma generalized LMM (GLMM) analysis with the KL divergence during centro-parietal SOs as the dependent variable and the fast spindle maturity component as a fixed effect, allowing for a random intercept for each participant. We further added the SO maturity score as a covariate. The GLMM revealed that only higher fast spindle maturity was associated with stronger modulation of development-specific fast centro-parietal sleep spindles during SOs (β = 0.50, t = 8.97, p < 0.001; see Figure 6C), while SO maturity was not significantly related to the modulation strength (β = 0.09, t = 1.28, p = 0.202). These results did also hold when controlling for age (Supplementary file 1v). In sum, our results suggest that developmental age differences in the manifestation of SO-spindle coupling are uniquely related to the degree to which the dominant, development-specific fast spindle type within an individual, that is, the sleep spindles that can be identified as peaks in the power spectrum (Aru et al., 2015; see Figure 1A), shares characteristics with adult-like fast sleep spindles.

Discussion

Although the synchronization of sleep spindles by the up peak of SOs and its functional significance has been recognized for decades, its development remains elusive (Muehlroth et al., 2019; Schreiner et al., 2021; Staresina et al., 2015; Steriade, 2006). Based on within-person detection of single events, we provide a detailed characterization of age-specific patterns of slow and fast sleep spindles, SOs, and their coupling in children aged 5–6 and 8–11 years, in adolescents aged 14–18 years, and in young adults aged 20–26 years. Specifically, in the child age groups, we noted that the predominant type of fast sleep spindles, as characterized by peaks in the power spectrum, is found in a frequency range slower than known from research in adults (canonical range, 12.5–16 Hz; Cox et al., 2017) and found also here in the adult sample (i.e., they manifest development-specifically). Surprisingly, the inspection of SO-spindle coupling patterns suggested synchronization being driven by spindles in the adult-like canonical fast spindle range—even in the younger age groups but notably less precisely timed during the SO cycle. Additional single event detection restricted to the canonical fast spindle range indeed revealed adult-like fast centro-parietal sleep spindle events in all age groups, although with only minor presence in the two child cohorts. Interrogation of coupling precision by means of PETHs confirmed that the coupling pattern found in adults, that is, reduced spindle occurrence during the down peak and increased sleep spindle likelihood precisely during the up state, can only be found for adult-like fast spindles and is less pronounced for the prevailing, development-specific fast spindles found in the child age groups.

To further corroborate these observations, we determined personalized measures for fast sleep spindle maturity based on the differences between the predominant, development-specific and the adult-like fast sleep spindles in terms of frequency, amplitude, and density. Indeed, we observed that higher fast spindle maturity (i.e., higher spindle maturity scores) was associated with a stronger SO-spindle modulation pattern of (i) decreased adult-like fast spindle power during the down state, (ii) increased adult-like fast spindle power during the up state, and (iii) increased canonical slow spindle power during the down peak. Most importantly, more advanced fast spindle maturity was exclusively linked to stronger modulation of development-specific fast centro-parietal spindles during SOs, over and above SO maturity. Hence, the present results provide evidence that the development of temporally precise SO-spindle coupling may specifically be linked to the maturation of fast sleep spindles.

Overall, our findings suggest that the temporally precise synchronization of sleep spindles during the up peak of SOs might not be an inherent feature of the thalamocortical system. Rather, SO-spindle coupling patterns differ systematically across the lifespan (Hahn et al., 2020; Helfrich et al., 2018; Joechner et al., 2021; Muehlroth et al., 2019), likely depending on age-specific anatomical and electrophysiological properties of the thalamocortical network.

Previous research in older adults suggested that age-related dispersion and imprecision of frontal SO-spindle coupling were related to atrophy in the prefrontal cortex and the thalamus—brain sites critically involved in the generation of mature SOs and spindles, respectively (Helfrich et al., 2018; Muehlroth et al., 2019). While these studies indirectly hypothesized that age-related changes in fast sleep spindle and SO features may contribute to alterations in the SO-spindle coupling pattern, we provide direct evidence that the development of the well-known coupling of sleep spindles to SO up peaks is uniquely linked to the emergence of an increasing number of spindles in the canonical fast sleep spindle frequency range. However, the exact anatomical and functional developments underlying the increased emergence of canonical fast spindles and their modulation by SOs remain to be elucidated. Furthermore, the question arises whether development-specific and canonical fast sleep spindles represent distinct representations of the same though developing generating network or whether they originate through different structures and connections.

Sleep spindles arise within well-described thalamic nuclei and thalamocortical circuits (Fernandez and Lüthi, 2020; Steriade, 1995). For one, the duration of hyperpolarization and the resulting length of hyperpolarization-rebound sequences in thalamocortical cells has been reported to particularly account for slower and faster spindle frequencies (Steriade, 2003; Steriade and Llinás, 1988). Hence, decreasing hyperpolarization in thalamocortical, and potentially reticular thalamic and cortical, cells may underly the global increase in the frequency of predominant sleep spindles across child and adolescent development (Campbell and Feinberg, 2016; Zhang et al., 2021). A relatively stronger age-related drop in hyperpolarization in topographically selected assemblies may lead to the expression of canonical fast spindles over posterior scalp electrodes. While the overall state of the brain determines the excitability of thalamocortical cells (Steriade and Llinás, 1988), it is still elusive which cellular or other developments could drive age-related alterations in thalamocortical hyperpolarizations.

Further, both thalamic nuclei and thalamocortical connectivity are refined across development (Steiner et al., 2020). This applies to structural and functional changes with studies indicating increased white matter and functional connectivity of thalamocortical tracts with older age—specifically with frontal, motor, and somatosensory cortical areas (Avery et al., 2022; Fair et al., 2010; Steiner et al., 2020). Previously, inter-individual differences in magnetic resonance imaging indicators of white matter properties, supposed to index myelin-dependent efficiency in signal transmission (Chanraud et al., 2010), were linked to the expression of spindle power and frequency (Mander et al., 2017; Piantoni et al., 2013b; Sanchez et al., 2020; Vien et al., 2019). Therefore, increased white matter integrity of certain thalamocortical projections may support the increase in sleep spindle frequency and higher numbers of fast sleep spindles with older age during development.

Importantly, accumulating findings support the conjecture that primarily canonical fast sleep spindles coalesce with the up peak of SOs, while slow sleep spindles are synchronized more towards the down peak (Bastian et al., 2022; Kurz et al., 2021; Mölle et al., 2011; Muehlroth et al., 2019). If merely the absolute frequency range of spindles would determine the coupling pattern, one might have simply expected to find a phase shift in coupling with older age and accompanying faster sleep spindles. However, in none of the age groups did we observe strong evidence indicating that sleep spindles outside of the canonical fast spindle frequency range or at other topographic locations couple to distinct phases of the SO. Rather than a synchronization of development-specific fast sleep spindles during the down state, we found a general pattern of lower spindle modulation in 5- to 6-year-olds where development-specific fast spindle frequency is the lowest. However, development-specific fast sleep spindles were not only lower in frequency but also occurred less often in children. Surprisingly though, the modulation of adult-like fast sleep spindles during SOs was well detectable across all age groups with this co-occurrence pattern becoming more precisely timed with SO up and down peaks with older age—despite huge age-related differences in their occurrence rate and the fact that adult-like fast sleep spindles showed very low density in both child cohorts (see Figure 1). Therefore, our data suggest that neither frequency nor density alone explains SO-spindle coupling patterns. Rather, in individuals increasingly capable of expressing adult-like, canonical fast sleep spindles, spindles at all frequencies are globally modulated more strongly and in a comparable fashion. However, it remains an open question how the distinct coalescence of canonical fast and slow spindles with SOs develops.

While sleep spindles can be solely initiated in intra-thalamic circuits and thus could co-occur with SOs merely by chance, corticothalamic input is one of the most potent mechanisms inducing spindles (Bonjean et al., 2011; Contreras and Steriade, 1995; Helfrich et al., 2018). The connectivity between the cortex and the thalamus is supposed to explain the formation of SO-spindle modulation in a way that the synchronous firing of cortical cells during the depolarizing SO up state can excite thalamic sleep spindle generation while the joint neuronal hyperpolarization during the down state terminates and inhibits spindles (Contreras and Steriade, 1995; Steriade, 2006). Hence, it is conceivable that enhanced SO-spindle coupling not only indexes the development of thalamocortical pathways (that may give rise to sleep spindles at faster frequencies) but also changes in corticothalamic connectivity allowing for increasingly efficient cortical control over spindle generation—specifically of canonical fast spindles (Contreras and Steriade, 1995).

While in our analyses SO maturity did not significantly explain SO-spindle coupling patterns over and above fast sleep spindle maturity, there is evidence indicating that SO characteristics are associated with the pattern of SO-spindle co-occurrence. For instance, a higher SO amplitude was associated with coupling strength (Kurz et al., 2021). Further, in older adults, a diminished directional influence of SOs was related to imprecise SO-spindle coupling (Helfrich et al., 2018). Therefore, we cannot exclude a potentially important contribution of the development and characteristics of SOs for SO-spindle coupling. It may just be that the prominent age-related differences in fast sleep spindles across childhood and adolescence may be relatively more influential and/or that SO features other than the ones examined here may be effective in shaping SO-spindle coupling.

On a functional level, the precisely timed synchronization of sleep spindles during SO up peaks is in particular discussed as important for memory consolidation (Maingret et al., 2016; Niknazar et al., 2015). In line with this, developmental differences in SO-spindle coupling from late childhood to adolescence have been linked to developmental enhancements in memory consolidation (García-Pérez et al., 2022; Hahn et al., 2020), implying less functional consolidation with less pronounced SO-spindle coupling (Helfrich et al., 2018; Muehlroth et al., 2019). Hence, besides the development of thalamocortical and corticothalamic networks, SO-spindle coupling might also reflect the functional maturation of hippocampal-neocortical networks (Cowan et al., 2020; Helfrich et al., 2018; Muehlroth et al., 2019). Yet, already children can show extraordinary levels of sleep-associated memory consolidation (e.g., Peiffer et al., 2020; Urbain et al., 2016; Wilhelm et al., 2013), raising the question of which features during sleep might support such high levels of memory consolidation at young age. Interestingly, in young children the development-specifically manifested SO-spindle coupling pattern was found to not yet be associated with memory consolidation while sleep spindles and SOs individually were (Joechner et al., 2021), implying that alternative features during sleep might compensate for less pronounced SO-spindle coupling (Wilhelm et al., 2012). However, these questions remain to be addressed in future longitudinal research.

Conclusion

Overall, our findings describe a unique relation between fast sleep spindle maturity and the pattern of SO-spindle coupling across childhood, adolescence, and young adulthood. While SOs provide optimal time windows for sleep spindles to arise, it is the ability to generate adult-like, canonical fast spindles that determines coupling strength and precision across child and adolescent development. Given the evidence for its generating role, our results implicate the maturation of specific thalamocortical circuits as the cornerstone of adult-like SO-spindle coupling patterns. Hence, our findings represent a promising starting point for future research addressing the precise relation between age-related changes in brain structure and function, the emergence of adult-like SO-spindle coupling, and cognitive development across childhood and adolescence.

Materials and methods

All analyses presented here are based on two previously published datasets: one cross-sectional cohort of 5- to 6-year-old children (Joechner et al., 2021) and one longitudinal sample of children, tested initially between ages 8–11 years (Hoedlmoser et al., 2014) and again around seven years later between ages 14–18 years (Hahn et al., 2019; Hahn et al., 2020) during adolescence. Please refer to the original studies for a detailed description of all inclusion and exclusion criteria and the experimental procedures, respectively. In addition, an unpublished cross-sectional sample of young adults aged between 20 and 26 years of age was analyzed.

Participants

For the cross-sectional child study, originally, 36 healthy, German-speaking 5- to 6-year-old pre-school children (19 female, Mage = 5 years, 9.53 months, SDage = 6.50 months) were recruited from the database of the Max Planck Institute for Human Development (MPIB), Berlin, Germany, and from daycare centers in Berlin, Germany. All 36 initially enrolled participants did not show any evidence of the use of medication, a personal or family history of mental and sleep disorders, learning disabilities, respiratory problems, and obesity. Of these, five participants did not complete the study, and another seven children were excluded due to technical issues (n = 4) during PSG or missing compliance with the study protocol (n = 3). Therefore, the same 24 5-to 6-year-old participants (13 female, Mage = 5 years, 10.71 months, SDage = 7.28 months) as in Joechner et al., 2021 are presented here.

The initial sample of the longitudinal cohort consisted of 63 healthy, Austrian children (T1, 28 female, Mage = 10 years, 1.17 months, SDage = 7.97 months; Hoedlmoser et al., 2014), recruited from public elementary schools in Salzburg, Austria. Approximately seven years later, 36 participants returned for a follow-up assessment (T2, 24 female, Mage = 16 years, 4.56 months, SDage = 8.76 months; Hahn et al., 2019; Hahn et al., 2020) during adolescence. At T1, none of the participants showed any signs of sleep or mental disorders, medication use, respiratory problems, and obesity. Two adolescents had to be excluded due to technical problems during PSG and data of one participant was not further analyzed because of insufficient amounts of NREM sleep stage 3 (N3; out of four nights, in only two nights N3 sleep was detected; together: 3.05%). Hence, we here present repeated-measures data for a total of 33 participants (23 female, MageT1 = 9 years, 11.70 months, SDageT1 = 8.35 months; MageT2 = 16 years, 4.91 months, SDageT2 = 9.06 months).

The cross-sectional sample of adults initially included 19 healthy, Austrian participants recruited at the University of Salzburg, Austria (15 female, Mage = 21 years, 8. 58 months, SDage = 19.49 months). One participant had to be excluded from analyses due to bad signal quality of the EEG data. Hence, we analyzed a final sample of 18 adult participants between the ages of 20 and 26 here (15 female, Mage = 21 years, 8.78 months, SDage = 20.04 months). Minor participants and their families received a gift (5- to 6-year-old and 8- to 11-year-old participants) and/or monetary compensation (the parents of the 5- to 6-year-olds and the 14- to 18-year-olds) for their study participation. Adult participants received either monetary compensation or study credit points for their participation.

All studies were designed in accordance with the Declaration of Helsinki and approved by the local ethics committee of either the MPIB, Germany (2018_5_Sleep_Conmem), or the University of Salzburg, Austria (EK-435 GZ:16/2014). For the 5- to 6-year-olds, legal guardians gave their written and minors gave their oral informed consent prior to study participation. For the longitudinal study, participants and their legal custodian provided written informed consent before entering the study at every test time point. Similarly, adults provided written informed consent. Given that participants were assessed during four distinct age intervals, the term ‘age group(s)’ is used to refer to 5- to 6-, 8- to 11-, 14- to 18-, and the 20- to 26-year-olds.

General procedure

Request a detailed protocol

Despite procedural differences, all three studies comprised two nights of ambulatory PSG for each test time point. While the first night served to familiarize participants with the PSG, a memory task was performed before and after the second night (an associative scene-object task for the cross-sectional child study and an associative word-pair task for both time points of the longitudinal study and the adult cross-sectional sample). Sleep was monitored in the habitual environment of each participant. For each 5- to 6-year-old participant, PSG recordings started and ended according to the individual bedtime. For the participants in the longitudinal sample, recordings were scheduled between 7:30–8:30 p.m. and 6:30 a.m. and were stopped after a maximum of 10 hr time in bed during childhood. During the follow-up assessment, time in bed was fixed to 8 hr between 11 p.m. to 7 a.m. For the cross-sectional adult sample, the procedure was exactly the same as for the longitudinal follow-up assessment. Given well-known first night effects in children (Scholle et al., 2003), only the second night was analyzed here.

Sleep EEG acquisition and analyses

Sleep recordings

Request a detailed protocol

For the youngest cohort, sleep was recorded using an ambulatory amplifier (SOMNOscreen plus, SOMNOmedics GmbH, Randersacker, Germany). A total of seven gold cup electrodes (Grass Technologies, Natus Europe GmbH, Planegg, Germany) were placed on the scalp for EEG recordings (F3, F4, C3, Cz, C4, Pz, Oz). EEG channels were recorded at a sampling rate of 128 Hz against the common online reference Cz. The signal of the AFz served as ground. Initial impedance values were kept below 6 kΩ. Additionally, two electrodes were placed at the left and right mastoids (A1, A2) for later re-referencing. Horizontal electrooculogram (EOG) was assessed bilaterally around the eyes and electromyogram (EMG) was recorded on the left and right musculus mentalis, referenced to one chin electrode. Furthermore, cardiac activity was monitored with two electrocardiogram (ECG) channels.

For the three older age groups, PSG signals were also collected ambulatory with a portable amplifier system (Varioport, Becker Meditec, Karlsruhe, Germany) at a sampling frequency of 512 Hz and against the common reference at Cz. For EEG recordings, 11 gold-plated electrodes (Grass Technologies, Natus Europe GmbH, Planegg, Germany; F3, Fz, F4, C3, Cz, C4, P3, Pz, P4, O1, O2) were placed along with A1 and A2 at the bilateral mastoids for offline re-referencing. Further, two horizontal and two vertical EOG channels, as well as two submental EMG channels, were recorded.

Pre-processing and sleep staging

Request a detailed protocol

Sleep was staged automatically (Somnolyzer 24x7, Koninklijke Philips N.V.; Eindhoven, The Netherlands) and visually controlled by an expert scorer according to the criteria of the American Academy of Sleep Medicine (AASM; see Supplementary file 1w for sleep architecture). Initial pre-processing was performed using BrainVision Analyzer 2.1 (Brain Products, Gilching, Germany). EEG channels were re-referenced offline against the average of A1 and A2 and filtered between 0.3–35 Hz. Further, all EEG data were resampled to 256 Hz. All subsequent analyses were conducted using Matlab R2016b (Mathworks Inc, Sherborn, MA) and the open-source toolbox Fieldtrip (Oostenveld et al., 2011). Firstly, bad EEG channels were rejected based on visual inspection. The remaining channels were then cleaned by applying an automatic artifact detection algorithm on 1 s segments (see Joechner et al., 2021; Muehlroth et al., 2019 for more details).

Sleep spindle and slow oscillation detection

Request a detailed protocol

Sleep spindles were detected during artifact-free NREM (N2 and N3) epochs at frontal (F3, F4) and centro-parietal channels (C3, C4, Cz, Pz) using an established algorithm (Klinzing et al., 2016; Mölle et al., 2011; Muehlroth et al., 2019) with individual amplitude thresholds. Frequency windows for spindle detection were defined based on two approaches: (i) An individualized approach, aiming at capturing the person- and development-specific dominant rhythm, and (ii) a fixed approach, targeting ‘adult-like’ fast sleep spindles in the canonical fast spindle range (12.5–16 Hz) that were shown to be coupled to SOs before, despite missing evidence for their strong presence in pre-school children (Joechner et al., 2021; see also Figure 1A and Figure 3). For the individualized approach, we defined the frequency bands of interest as the participant-specific peak frequency in the averaged frontal (F3, F4) or centro-parietal (C3, Cz, C4, Pz) background-corrected NREM power spectra (9–16 Hz) ± 1.5 Hz (Mölle et al., 2011; Ujma et al., 2015; see Figure 1A for peak distributions). Therefore, NREM power spectra were calculated within participants for averaged frontal and centro-parietal electrodes between 9–16 Hz by applying a Fast Fourier Transform (FFT) and using a Hanning taper on 5 s epochs. To restrict the search space for peak detection to dominant rhythmic activity (Aru et al., 2015; Kosciessa et al., 2020), we then modeled the background spectrum and subtracted it from the original power spectrum. Assuming that the EEG background spectrum follows an A*f-a distribution (Buzsáki and Mizuseki, 2014; He et al., 2010), background spectra were estimated by fitting the original power spectrum linearly in the log(power)-log(frequency) space employing robust regression (Kosciessa et al., 2020). Frontal and centro-parietal peaks were finally detected in the background-corrected power spectra by a classical search for maxima combined with a first derivative approach (Grandy et al., 2013). For the fixed approach, frequency bands were restricted to 12.5–16 Hz in every individual—representing the canonical range for fast sleep spindle extraction in adults—and detection was focused on averaged centro-parietal electrodes only to capture adult-like fast sleep spindles (C3, Cz, C4, Pz; Cox et al., 2017). For detection of spindle events, EEG data were first filtered in the respective frequency bands using a 6th order two-pass Butterworth filter (separately for individually identified frontal and centro-parietal, as well as adult-like fast centro-parietal spindles). Then, the root mean square (RMS) was calculated using a moving window of 0.2 s and smoothed with a moving average of 0.2 s. Finally, a sleep spindle was detected whenever the envelope of the smoothed RMS signal exceeded its mean by 1.5 SD of the filtered signal for 0.5–3 s. Sleep spindles with boundaries within a distance between 0–0.25 s were merged as long as the resulting event was shorter than 3 s (see, e.g., Mölle et al., 2011; Muehlroth et al., 2019, for similar methods; see Figure 1—figure supplements 1 and 2 for the average EEG signal time-locked to all different spindle types and for examples of how individually identified (i.e., development-specific) and adult-like fast centro-parietal spindles manifested in the raw EEG).

SO detection was based on an algorithm with individually adapted amplitude thresholds (Mölle et al., 2002; Muehlroth et al., 2019) and performed for all NREM epochs. First, the EEG signal was filtered between 0.2 and 4 Hz using a two-pass Butterworth filter of 6th order. Subsequently, zero-crossings were marked to identify positive and negative half-waves. Pairs of negative and succeeding positive half-waves were considered a potential SO if their frequency was 0.5–1 Hz. Only putative SOs with a peak-to-peak amplitude of 1.25 times the mean peak-to-peak amplitude of all tagged SOs and a negative amplitude of 1.25 times the average negative amplitude of all SOs that did not include artifact segments were kept for the following analyses.

Temporal association between sleep spindles and slow oscillations

Request a detailed protocol

To assess the temporal association between sleep spindles and SOs, we employed two different approaches following previous reports in developmental samples (Joechner et al., 2021; Muehlroth et al., 2019).

Time-frequency analyses
Request a detailed protocol

On the one hand, we examined power modulations during SOs between 5–20 Hz within intervals of ± 1.2 s around the down peak of SOs. For this, we detected artifact-free NREM epochs containing SOs (down peak ± 3 s) and equally long, randomly chosen segments without SOs. Time-frequency representations (Figure 3) were then calculated for trials with and without SOs using a Morlet wavelet transformation (12 cycles) in steps of 1 Hz and 0.002 s. The resulting time-frequency pattern of trials with and without SOs was contrasted for every participant using independent sample t-tests. Subsequent group-level analyses were conducted separately for the cross-sectional cohorts and the two cohorts from the longitudinal sample. Within these four age groups, the t-maps were then contrasted against zero using a cluster-based permutation test (Maris and Oostenveld, 2007; two-sided, critical alpha-level α=0.05; note for the ease of comparability, p-values for all cluster-based permutation tests were multiplied by 2 and thus values below 0.05 were considered significant) with 5000 permutations in a time window between –1.2 to +1.2 s around the SO down peak.

Temporal co-occurrence of sleep spindle events with slow oscillations
Request a detailed protocol

On the other hand, we investigated the temporal co-occurrence of individually identified frontal and centro-parietal and adult-like fast sleep spindle events with SOs. In a first step, we calculated the general co-occurrence rate between sleep spindles and SOs on a broad time scale by determining the percentage of spindle event centers that occurred within an interval of ± 1.2 s around SO down peaks during NREM sleep, relative to all spindles during NREM sleep. We also repeated this analysis vice versa for SO down peaks occurring within an interval of ± 1.2 s around spindle centers (see Figure 4—figure supplements 3 and 4 and Supplementary file 1x–ac).

To specify the temporal relation between sleep spindle and SO co-occurrence on a finer temporal scale, in a second step, we calculated PETHs (Figure 4). Therefore, the interval of ± 1.2 s around a SO down peak was partitioned into 100 ms bins and the proportion of sleep spindle centers occurring within each time bin was assessed. The occurrence rates within a bin were subsequently normalized by the total counts of sleep spindles occurring during the complete respective SO down peak ± 1.2 s interval and multiplied by 100. To determine whether spindle activity was modulated during SOs differently from what would be expected by chance, we created a surrogate comparison distribution for every participant during every test time point. The original percentage frequency distribution was randomly shuffled 1000 times and averaged across all permutations. The resulting surrogate distributions were then compared against the original distributions using dependent sample t-tests. To account for the multiple comparisons, a cluster-based permutation test with 5000 permutations was implemented (Maris and Oostenveld, 2007; two-sided, critical alpha-level α=0.05, p-values for all cluster-based permutation tests were multiplied by 2 and thus values below 0.05 were considered significant). All analyses were conducted separately for individually identified sleep spindles at frontal (averaged over F3, F4) and centro-parietal (averaged over C3, Cz, C4, Pz) electrodes and for adult-like fast spindles detected at centro-parietal (averaged over C3, Cz, C4, Pz) derivations. For a control analysis on the influence of the number of events on PETH patterns, we 100 times randomly drew half the number of adult-like fast sleep spindles within a participant and calculated co-occurrence rates. We than averaged across these 100 surrogates and recalculated the PETHs.

To quantify the modulation strength of sleep spindle occurrence during SOs, for every participant, we calculated the KL divergence between the percentage frequency distribution (p) and its surrogate (q) used for the PETHs. The KL divergence is a measure rooted in information theory that describes the amount of information loss if one distribution was approximated by the other (Joyce, 2011). This measure is the basis for several commonly used methods to determine phase-amplitude coupling and was calculated in the following way:

DKL(p||q)=i=1N(p(xi)lnp(xi)q(xi))

The higher the value, the more two distributions deviate from each other. Hence, higher values indicate that the actual percentage frequency distribution of the PETHs deviated from the surrogate distribution, that is, the more likely sleep spindles were concentrated around specific phases of the underlying SO.

Statistical analyses

Age differences in sleep spindles and slow oscillations

Request a detailed protocol

Age-related comparisons between sleep spindle and SO parameters were conducted using LMMs with restricted maximum likelihood variance estimation (REML) and the bobyqa optimizer from the NLopt library (Powell, 2009). LMMs were implemented in R 4.0.3 (R Development Core Team, 2020) using Rstudio Version 1.1.383 and the lme4 package (Bates et al., 2015). Given that our data consist of both, purely cross-sectional as well as repeated measures, participants were included as random effects to account for the nonindependence when dealing with repeated measures. Fixed effects were most of the time ‘age group’ (i.e., 5- to 6-, 8- to 11-, 14- to 18-, 20- to 26-year-olds) and ‘topography’ (e.g., frontal, central, occipital) and specified based on our research questions. The effects of categorical predictors were set up as sum-coded factors. Overall, only models with a random intercept for participants converged (but not with a random slope). Results were summarized by F-statistics and p-values based on Type III sums of squares estimation and the Satterthwaite’s method provided by lmerTest (Bates et al., 2015; Kuznetsova et al., 2017). Post-hoc pairwise comparisons were conducted using the emmeans package (Lenth, 2021). Degrees of freedom (df) were calculated applying the Satterthwaite’s method (Giesbrecht and Burns, 1985; Kuznetsova et al., 2017) and p-values were Bonferroni corrected (padj; Bland and Altman, 1995).

Association between sleep spindle and slow oscillation maturity with slow oscillation-spindle coupling

Request a detailed protocol

To assess the association between sleep spindle and SO maturity and SO-spindle coupling (for an exact definition of the variables, see the results section), we conducted two analyses:

First, a PLSC (Krishnan et al., 2011) was calculated between indicators of sleep spindle amplitude, frequency, and density maturity and the t-maps of power modulations during SOs (compared to non-SO trials, time-frequency maps) across all age groups. PLS methods are multivariate tools to extract commonalities between two data sets using singular value decomposition. PLSC specifically analyzes the association between two data sets by deriving pairs of singular vectors that cover the maximal covariance between two matrices (Krishnan et al., 2011). Singular vector pairs are ordered by the amount of covariance they contribute to the association as reflected by their corresponding singular values. The components of the singular vectors represent weights (also called saliences) that define how each element of the data sets contributes to a given association. Ultimately, the projection of pairs of saliences onto their original data matrices results in pairs of latent variables that capture the maximal amount of common information between the two data sets (e.g., in our case a latent time-frequency and latent spindle maturity variable). Therefore, PLSC is ideally suited to identify time-frequency patterns associated with a specific spindle maturity profile. The number of impactful singular vector pairs (i.e., those that account for a significant amount of covariance) was identified by a permutation test with 5000 permutations based on the singular values. For each significant vector pair, reliability of the weights of each time-frequency value, defining the SO-spindle coupling pattern associated with a spindle maturity pattern, was determined using bootstrap ratios (BSR). BSRs represent the ratio of the weights of each time-frequency value and their bootstrap standard errors based on 5000 samples. BSRs are akin to Z-scores, hence BSRs higher than 1.96 and lower than –1.96 were considered stable (Krishnan et al., 2011). The spindle maturity profile was represented as the correlation between the latent time-frequency variable and the three raw spindle maturity variables. These values are comparable to the spindle maturity weights and indicate whether the association between the time-frequency pattern and individual spindle maturity variables is in the same or different direction (McIntosh and Lobaugh, 2004). Stability of these correlation patterns was determined by their bootstrap estimated 95% confidence intervals (McIntosh and Lobaugh, 2004). Note, despite our awareness of different dependencies between the age groups, 5- to 6-, 8- to 11-, 14- to 18-, and 20- to 26-year-olds were considered for this analysis.

Second, GLMMs were used to examine the relation between the modulation of sleep spindles during SOs (as captured by the KL divergence of the PETHs) with spindle and SO maturity across all age groups. We defined the first component of a PCA between indicators of sleep spindle amplitude, frequency, and density maturity across the complete data set (across all age groups) as a general spindle maturity factor. We then aimed to associate the KL divergence values, reflecting sleep spindle modulation strength during SOs, with the general spindle maturity component. For this, we set up log-linked gamma GLMMs with the KL divergence as the dependent variable and with the spindle maturity component as a fixed factor, allowing for random intercepts per participant. In addition, an indicator of SO maturity was added as a covariate. We used the bobyqa optimizer from the minqa package (Bates et al., 2014).

Data availability

All data and code to reproduce the present analyses and result figures are publicly available through the Open Science Framework (https://osf.io/2u6ec/).

The following data sets were generated
    1. Joechner A-K
    2. Werkle-Bergner M
    (2022) Open Science Framework
    ID 2u6ec. Sleep spindle maturity promotes slow oscillation-spindle coupling across child and adolescent development.

References

    1. Joyce JM
    (2011)
    International Encyclopedia of Statistical Science
    Kullback-Leibler divergence, International Encyclopedia of Statistical Science, Springer, 10.1007/978-3-642-04898-2.
  1. Book
    1. Powell MJD
    (2009)
    The BOBYQA algorithm for bound constrained optimization without derivatives
    Department of Applied Mathematics and Theoretical Physics.
  2. Software
    1. R Development Core Team
    (2020) R: A language and environment for statistical computing
    R Foundation for Statistical Computing, Vienna, Austria.

Decision letter

  1. Laura L Colgin
    Senior and Reviewing Editor; University of Texas at Austin, United States
  2. Hong-Viet Ngo
    Reviewer; University of Essex, United Kingdom

Our editorial process produces two outputs: (i) public reviews designed to be posted alongside the preprint for the benefit of readers; (ii) feedback on the manuscript for the authors, including requests for revisions, shown below. We also include an acceptance summary that explains what the editors found interesting or important about the work.

Decision letter after peer review:

Thank you for submitting your article "Sleep spindle maturation enhances slow oscillation-spindle coupling" for consideration by eLife. Your article has been reviewed by 3 peer reviewers, and the evaluation has been overseen by Laura Colgin as the Senior Editor. The following individual involved in review of your submission has agreed to reveal their identity: Hong-Viet Ngo (Reviewer #3).

The reviewers have discussed their reviews with one another, and the Reviewing Editor has drafted this to help you prepare a revised submission.

Essential revisions:

1) A major limitation of the work is that analogous analyses were not performed on sleep data recorded from adult subjects. Reviewers strongly encourage including results from adults in the resubmitted manuscript. However, if it is not feasible to complete such analyses within the time allowed for revisions, the expectation would be that supporting data from adults would be submitted to eLife in the future as a follow-up to the present study (e.g., submit the new results from adults to eLife as a Research Advance in the future see: https://reviewer.elifesciences.org/author-guide/types.

2) Additional analyses and revisions suggested to improve clarity and maximize impact of the work are provided in the individual reviews below.

Reviewer #1 (Recommendations for the authors):

1. An unfortunate and I think unnecessary limitation of this manuscript is that a sample of adults is missing. This would be highly desirable to show how the very complex analyses and transformations compare to previously reported analyses in young healthy adults. It would make interpreting the data much easier and there are plenty such samples available either in the author's own labs or from their collaborators throughout Germany, Austria and Switzerland. I would urge the authors to add such data to the manuscript.

2. One other concern regards the spindle maturation index. The authors report two fast spindle peaks: one age adapted and one canonical. What is the evidence that the "age-adapted" fast spindles are indeed fast spindles? How do we know that they are generated by the same thalamocortical circuits? I believe this is crucial for the argument of the authors. Since the "age-adapted" fast spindles are not modulated by the SO this may indicate that they are not the same phenomenon. Clearly, this cannot be shown conclusively in this paper, but I would urge the authors to more stringently test this, to then thoroughly discuss this and to compare the features of these oscillations more thoroughly.

3. Since the second time point in the longitudinal sample was assessed during puberty, I was wondering, if the authors took sex into account. It seems likely that sex hormones could affect their findings.

4. Since the authors find less canonical spindles in younger participants, could the effect of imprecise spindle to SO coupling shown in Figure 4 be due to noise? To check this the authors could randomly choose a subset of spindles for the adolescents and see how precision compares then.

5. Related to this. The findings that spindle maturation is related to SO-spindle coupling seems like double dipping. First, the authors identify age related differences in two measures and then they find a relationship between them. I think it would be a stronger finding, if this relationship was shown within each age group individually. However, I may have misunderstood this approach.

6. Generally, the analyses are a bit hard to understand, since for example a lot of transformations are being performed before analysis. I would suggest adding some more descriptive figures to the manuscript that allow judging the pipeline from raw data to inference. For example, I would like to see averages (for all subjects) and grand averages (per age group) of the different types of detected spindles. I would also like to see the spindle band-filtered EEG signal (for each type of spindle) time-locked to the slow-oscillation down-state. The authors should also add figures showing the averaged (and SD) spectral analyses that were used to identify the different spindle peaks. Also the individual spectra could be shown larger in the supplement.

7. Can the authors analyse the phase-amplitude-coupling of the SOs and canonical spindles per age group to extract the phase angle? There seems to be a phase progression from younger to older, which I find quite interesting.

8. The findings are of course the result of data exploration. This should be clearly acknowledged. It might be helpful to add some robustness analyses to show that the findings are not due to specific choices made during the exploration process.

9. The time-frequency plots time-locked to SOs do not seem to show a meaningful comparison. This is usually done by comparing two conditions rather than to a baseline. I am not sure we can learn much from them, if there is just a huge cluster across the whole spectrum and time.

10. For better readability: please do not abbreviate spindles.

Typos: "However, in neither age group, we did" (also: AFAIK "neither" is only used for two alternatives).

Reviewer #2 (Recommendations for the authors):

P3. Line 5: memories are not "transferred" to the cortex but strengthened in the cortex (since already encoded in the cortex).

After reading the whole article it is clear why the authors use "canonical" for the "normal" spindles, but it would be easier for readers if the authors the first time when they use the word, explain why and define it.

Reviewer #3 (Recommendations for the authors):

Title: The title is rather unspecific with respect the age range and could be specified further.

Figures: Almost all plots showing individual data points, means and raincloud plots are very hard to interpret and could need a small overhaul. In particular Figure 2, does not allow a visual assessment of the differences between cortical areas.

Page 1, line 4: For consistency, the authors could also add the frequency range of SOs here.

Page 2, line 11: Here it is stated that how the coupling develops is an open question. However, at the end of the next paragraph, a previous publication is reported that shows how the coupling improves across childhood. Is this open question still valid then?

Moreover, throughout the introduction, the authors repeat this open question several times with slights variations in the phrasing. I wonder if this is necessary or if the open question can be asked centrally before the final paragraph of the introduction.

Page 6, line 4: Please see my comment in the public review. I think the assessment of average frequency of spindle events is redundant or even circular as they are based on the power spectra.

Page 6, line 18: The introduction of development-specific feels unnecessary and to some degree arbitrary. At least for me, it only led to overly complicated/cluttered sentences. Is this term really required or can it be omitted while the main message is conveyed properly.

Page 10, Figure 1: It would help the reader if the age ranges are illustrated in subpanel A. Moreover, as stated above, symbols for the different spindle types and the marginal means are very hard to distinguish. Also, I was at first not aware that the adult-like spindles even had a rain cloud. Perhaps the authors can rearrange/resize the figure.

Related to subpanel A, its known that adults sometimes do not show slow spindle peaks or shows both a slow and fast spindle peak at one electrode (e.g., Mölle et al., 2011). At least the former seems to be the same here. How did the authors proceed if that was the case? I was wondering if it would be more informative to sort the individual subjects peak frequency per age group.

Page 14, line 4: What do the authors mean by "more precise"? The size of the cluster or its location with respect to the SO up-state? Is it really possible to make conclusion about the precision using a TFR-based analysis affected by temporal and frequency smearing?

Page 14, Figure 2. A comparison between SO vs. SO-free intervals includes a strong difference in overall spectral power, hence the wide significant cluster. The authors could consider decreasing their α-value further to tease out clusters more clearly. Furthermore, the dashed window highlighting the spindle range is not really visible.

Page 16, Figure 4: Again, this is reflecting my personal opinion, but this the central result and it somehow feels like it does not get enough attention. Anyhow, what does the horizontal jiggly line represent? It is also stated that the empirical PETHs are compared to a surrogate distribution, however, is that really necessary? The authors have SO-free intervals that could be used for a cluster-based permutation test as well.

Page 19, Figure 6: Can the author please explain what the main message of subpanel A is? This somehow eludes me. Also, how come there is not separation into the different age groups?

https://doi.org/10.7554/eLife.83565.sa1

Author response

Essential revisions:

Reviewer #1 (Recommendations for the authors):

1. An unfortunate and I think unnecessary limitation of this manuscript is that a sample of adults is missing. This would be highly desirable to show how the very complex analyses and transformations compare to previously reported analyses in young healthy adults. It would make interpreting the data much easier and there are plenty such samples available either in the author's own labs or from their collaborators throughout Germany, Austria and Switzerland. I would urge the authors to add such data to the manuscript.

We now included the analysis of a sample of adults that were recorded with the same ambulatory amplifier and setup as the longitudinal sample.

In summary, the analyses on the adult age group showed similar results as in the age group of 14- to 18-year-olds. For sleep spindle analyses, in line with the literature (Cox et al., 2017; Ujma et al., 2015), the comparison between individually identified centro-parietal spindles (in the child age groups, development-specifically manifested) and those detected based on fixed canonical criteria (12.5–16 Hz) nicely demonstrates that there were no differences in any of the assessed features, making us even more confident to call these sleep spindles adult-like fast sleep spindles. Slow oscillation-spindle co-occurrence analyses revealed the expected pattern of increased occurrence of fast sleep spindle activity in the canonical frequency range (≈ 12.5–16 Hz) during the slow oscillation up peak and decreased activity during the down peak, highly similar to the results from the 14- to 18-year-olds. Lastly, taking the adult sample into account for the analyses on how sleep spindle and slow oscillation maturity relate to slow oscillation-spindle coupling patterns across all age groups revealed comparable results as before. Namely that a higher similarity between dominant, development-specific fast sleep spindles and adult-like canonical fast spindle features was associated with a more pronounced slow oscillation-spindle coupling pattern.

2. One other concern regards the spindle maturation index. The authors report two fast spindle peaks: one age adapted and one canonical. What is the evidence that the "age-adapted" fast spindles are indeed fast spindles? How do we know that they are generated by the same thalamocortical circuits? I believe this is crucial for the argument of the authors. Since the "age-adapted" fast spindles are not modulated by the SO this may indicate that they are not the same phenomenon. Clearly, this cannot be shown conclusively in this paper, but I would urge the authors to more stringently test this, to then thoroughly discuss this and to compare the features of these oscillations more thoroughly.

The question on the nature of distinct surface representations of sleep spindles is in our view one of the crucial questions in the field and also the central question arising from our results. Therefore, we already discuss this on page 28, lines 2–5.

To clarify, within a given participant we did not find separable “fast sleep spindle peaks” for development-specific and canonical fast spindle ranges, respectively. On the contrary, the (only) spectral peak for sleep spindles in centro-parietal areas (where fast sleep spindles usually dominate) was markedly below the frequency of canonical fast sleep spindles for the majority of younger children (see Figure 1A). As described under 2.1 “Dominant fast cento-parietal sleep spindles become more prevalent and increasingly resemble canonical fast sleep spindles with older age”, we considered sleep spindles in centro-parietal areas, detected based on the individual peak frequency, as fast based on a combination of frequency (faster than dominant frontal sleep spindles) and topography (centro-parietal derivations). Traditionally, Gibbs and Gibbs (1950) have defined slow and fast spindles based on these criteria and such a definition is also commonly employed nowadays (Anderer et al., 2001; Cox et al., 2017; Mölle et al., 2011; Ujma et al., 2015; for review see: De Gennaro and Ferrara, 2003; Fernandez and Lüthi, 2019). Since we define slow frontal and (development-specific) fast sleep spindles based on the respective peak frequency within a participant, we are confident to capture comparable phenomena with this approach, namely the dominant slow and fast spindle types within a participant at a given age. This is in our view a crucial feature for age-fair comparisons. However, even in the absence of a peak in the canonical fast spindle frequency range (as for the younger children, while for older children, adolescents, and young adults, the individual centroparietal peak frequency was already within or very close to the adult-like, canonical fast spindle frequency range), we could additionally find a low number of adult-like, canonical fast sleep spindles—based on detection of individual oscillatory events in the canonical fast spindle range. While the sleep spindles detected based on the peak frequency and those based on the fixed, adult-like, canonical fast frequency range likely capture similar events for older children, adolescents, and young adults, it remains to be elucidated whether the dominant, development-specific fast spindles in young children are a slower version of canonical fast spindles, or whether they represent distinct phenomenon. As discussed on page 28, specifically the question that needs to be answered is:

“…whether development-specific and canonical fast sleep spindles represent distinct representations of the same though developing generating network or whether they originate through different structures and connections.”.

While we agree on the importance of this question, we do not see a possibility how to test this further than already done with the present data. In our view, testing the question of whether development-specific and adult-like fast sleep spindles arise from similar or distinct structures and networks would at least require high density polysomnography for source reconstructions or combined sleep polysomnography and structural magnetic resonance data in human participants—ideally in longitudinal studies.

3. Since the second time point in the longitudinal sample was assessed during puberty, I was wondering, if the authors took sex into account. It seems likely that sex hormones could affect their findings.

We agree with the reviewer that this is an important question. Indeed, sex hormones could potentially affect sleep-specific rhythmic neural activity during maturation. However, this was not a question in the context of the present study. Hence, we did not explicitly collect any information on sex steroids like e.g., estrogens and androgens. Sex differences in sleep have often been reported and associated with a variety of potential hormones like estradiol, melatonin, growth hormone and prolactin release (Franco et al., 2020; Mong and Cusmano, 2016; for review see, e.g., Attarian and Viola-Saltzman, 2020; Fernandez and Lüthi, 2020). However, the research on sexual dimorphism in sleep is still in its infancy (Mong and Cusmano, 2016) and sex differences do not (yet) allow inferences on which hormones could account for given differences. This applies even more for sexual dimorphism across maturation, where the current literature is very inconsistent (Franco et al., 2020). Since we did not have a specific hypothesis about sex differences in the context of the present study, we did not take sex into account in our analyses. Based on the previous points and given an unbalanced number of males and females in the longitudinal sample (f = 23, m = 10) and the newly added adult sample (f = 15, m=3), that we believe would bias statistical results, we would like to remain with the decision to not analyze sex differences in the context of the present paper.

However, the reviewer raises an important point. Therefore, we screened for sex differences in the course of the revision. We did not observe marked differences between females and males for our reported results (see Author response image 1 and 2 for sleep spindle and slow oscillation features grouped by sex as examples).

Author response image 1
Sleep spindle (A) frequency (B) density, and (C) amplitude for slow frontal, development specific fast centro-parietal, and adult-like fast sleep spindles grouped by age group and sex.

Asterisks represent the mean. Individual data points represent individual participants. Note, specifically for the adolescent and the adult sample, the number of females and males was highly unbalanced.

Author response image 2
Slow oscillation (A) frequency (B) density, and (C) amplitude grouped by age group and sex.

Asterisks represent the mean. Individual data points represent individual participants. Note, specifically for the adolescent and the adult sample, the number of females and males was highly unbalanced.

4. Since the authors find less canonical spindles in younger participants, could the effect of imprecise spindle to SO coupling shown in Figure 4 be due to noise? To check this the authors could randomly choose a subset of spindles for the adolescents and see how precision compares then.

Thank you for this question. Indeed, a differing total number of detected events (sleep spindle and slow oscillation events) could bias certain coupling and co-occurrence analyses. Being aware of this issue in between-person and age-comparative analyses, we already normalized occurrence rates of spindles within 100 ms time bins during a slow oscillation by the overall number of sleep spindles co-occurring with that slow oscillation per participant when creating the peri-event time histograms (PETH) instead of using absolute values. Further, we also discuss whether a lower number of sleep spindles might drive age differences in coupling patterns on page 29, lines 15–22:

“However, development-specific fast sleep spindles were not only lower in frequency but also occurred less often in children. Surprisingly though, the modulation of adult-like fast sleep spindles during SOs was well detectable across all age groups with this co-occurrence pattern becoming more precisely timed with SO up and down peaks with older age—despite huge age-related differences in their occurrence rate and the fact that adult-like fast sleep spindles showed very low density in both child cohorts (see Figure 1). Therefore, our data suggest that neither frequency nor density alone explains SO-spindle coupling patterns.”

A lower base number of spindles should only affect our PETH analyses if the selected subset of spindles would be biased on a temporal scale, but not so much if missing at random. In order to back this up, we actually ran the PETH analyses for adolescents also based on the suggested procedure: for every participant, we randomly selected half the number of detected adult-like fast sleep spindles 100 times and calculated the occurrence rates per time bin. We than averaged across these 100 surrogates and recalculated the PETHs. We observed a very similar pattern for the analyses with half the number of adult-like fast spindles as compared to the original analyses. Hence, we are quite convinced that the age-related differences in our PETH analyses are not based on noise caused by differing numbers of events. We have added this control analysis as Figure 4—figure supplement 2.

Further, it should be noted that while there was a significantly lower number of both dominant, development-specific and adult-like fast sleep spindles in younger compared to older participants, interestingly, when looking at how many percent of those co-occurred with slow oscillations, there were only slight differences in the percentage of fast sleep spindles that did occur within the time-window of a slow oscillation between the age groups. These differences were not consistent in terms of higher co-occurrence rates for older than younger participants (or vice versa), but differed depending on the slow oscillation topography and sleep spindle type, e.g., for adult-like fast sleep spindles, the 5- to 6-yearolds showed the highest slow oscillation-spindle co-occurrence with centro-parietal slow oscillations, while there were no differences between age groups in co-occurrence rates for development-specific fast spindles with frontal and centro-parietal slow oscillations (see Supplementary files 1x–ac and Figure 4—figure supplements 3 and 4). Hence, the number of events does not easily map on their temporal occurrence.

5. Related to this. The findings that spindle maturation is related to SO-spindle coupling seems like double dipping. First, the authors identify age related differences in two measures and then they find a relationship between them. I think it would be a stronger finding, if this relationship was shown within each age group individually. However, I may have misunderstood this approach.

We thank Reviewer 1 for sharing her/his concerns about associating age-related features in sleep spindles and slow oscillations with our two measures of slow oscillation-spindle coupling. Since we are not sure, whether we understood these concerns correctly, we firstly want to clarify our approach:

Within every participant, we detected fast sleep spindles based on the individual peak frequency (development-specific) and additionally using a fixed frequency band in the canonical, i.e., adult-like, fast sleep spindle frequency range (12.5–16 Hz). Since the differences between sleep spindles detected with these two methods were bigger in younger and smaller in older participants, we quantified this observation by computing the distance between these sleep spindles (e.g., average frequency of adult-like fast sleep spindles minus average frequency of development-specific fast centro-parietal sleep spindles, average amplitude of adult-like fast sleep spindles minus average amplitude of development-specific fast centro-parietal sleep spindles, average density of adult-like fast sleep spindles minus average density of development-specific fast centro-parietal sleep spindles ). As a result, for every participant within each age group, we received a numerical value that describes how close adult-like fast and dominant development-specific fast sleep spindles are within a participant at a given age. This within-participant difference measure varied with age in a way that the differences were smaller at older ages—therefore we called these difference scores fast sleep spindle maturity scores (see also Figure 5 A–C). We then aggregated these values for frequency, amplitude, and density using Principal Component Analysis which gave us what we call the fast sleep spindle maturity component (see Figure 5 E). In a similar way, we quantified the maturity of slow oscillations within every participant within each age group by the (dis-)similarity between frontal and centro-parietal amplitude. This within-participant difference score also varied with age in way that older participants showed a higher difference score (see Figure 5 D). Therefore, we termed it slow oscillation maturity score. To sum up, the maturity scores reflect features within a participant at a given age that each varied with age.

We then related these maturity scores with a quantification of the slow oscillation-spindle coupling pattern per participant at a given age (see Figure 5 F). Hence, we related within participant measures that presumably reflect the maturity of sleep spindles, slow oscillations, and their coupling. Importantly, while it might be possible that measures that individually are related to age in a comparable fashion could also be associated, this does not automatically imply that these measures also covary.

Our finding of a significant association between the sleep spindle maturity score(s) and the coupling patterns across our whole sample provides evidence that these features indeed covaried in our sample. This implies that there might be a relation between the manifestation of sleep spindles and the pattern of slow oscillation-spindle coupling across our studied age range.

Further, the covariation suggests that there could also be a common source of this variance related to the age of participants. Importantly though, including both sleep spindle and slow oscillation maturity measures in the generalized linear mixed-effects models did only reveal a significant relation between our measures of sleep spindle maturity and modulation strength but not with slow oscillation maturity, already indicating that age is not the only underlying source driving this relation.

As already discussed in Lindenberger and Pötter (1998), a relation between two age-related variables would be spurious if this relation is exclusively driven by age, meaning that the related variables do not share any other variance over and above that shared with age—they are completely uncorrelated when controlling for age (see Lindenbeger and Pötter, 1998, Figure 1). However, if a relation between these variables remains after partializing out age, one can be confident that this relation is influenced by common variance that is orthogonal to that of age (Lindenberger and Pötter, 1998, Figure 2). Indeed, taking age into account in our generalized linear mixed-effects models, the results reported in the manuscript and Supplementary file 1 did hold. Sleep spindle maturity was still significantly related to centroparietal (β = 0.25, t = 3.00, p = .003) and frontal slow oscillation-spindle modulation strength (β = 0.25, t = 3.11, p = .002), while the slow oscillation maturity score was not (centro-parietal slow oscillation-spindle modulation strength: β = 0.002, t = 0.03, p = .977; frontal slow oscillation-spindle modulation strength: β = 0.04, t = 0.64, p = .526). In addition, age was related to centro-parietal (β = .32, t = 3.76, p < .001) and frontal slow oscillation-spindle modulation strength (β = .18, t = 2.23, p = .026). Overall, this implies that the relation between sleep spindle maturity and slow oscillation-spindle coupling strength is not merely driven by age.

However, given its cross-sectional nature our results do neither mean that the presented pattern would hold for within-person changes across the studied age range (Lindenberger et al., 2011), nor do they imply any causality, e.g., of sleep spindle changes causing changes in coupling patterns.

Further, analyzing the relationship between spindle and slow oscillation maturity and the coupling patterns within each age group could theoretically be done. However, the following points have to be considered:

1) given the small sample sizes within each age group statistical power would be decreased

2) given the different age ranges variances likely differs between age groups

and connected to the previous points, one would assume that there is measurement invariance between the different age groups, i.e., the association is supposed to be exactly the same in 5- to 6-year-olds as in 14- to 18-year-olds which is likely not the case.

3) Therefore, analyses within age groups are subject to several influences that also bias results.

Hence, overall, we are confident that our approach is a valid and justified way to examine whether sleep spindle and slow oscillation features across different ages are related to their coupling patterns. We have added the analyses controlling for age to our Supplementary file 1 as Supplementary file 1v.

6. Generally, the analyses are a bit hard to understand, since for example a lot of transformations are being performed before analysis. I would suggest adding some more descriptive figures to the manuscript that allow judging the pipeline from raw data to inference. For example, I would like to see averages (for all subjects) and grand averages (per age group) of the different types of detected spindles. I would also like to see the spindle band-filtered EEG signal (for each type of spindle) time-locked to the slow-oscillation down-state. The authors should also add figures showing the averaged (and SD) spectral analyses that were used to identify the different spindle peaks. Also the individual spectra could be shown larger in the supplement.

We thank Reviewer 1 for her/his suggestions and agree, that visualizing the different sleep spindle types is important. We might not have stated this clearly enough in the original manuscript:

“[…] see Supplementary Figure 1 for an example raw EEG trace with development specific and adult-like fast spindles and Supplementary Figure 2 for examples of average detected events”, but the original Supplementary file already included the averaged electroencephalographic (EEG) signal time-locked to the center point of (1) individually identified development-specific fast centro-parietal and (2) adult-like fast sleep spindles for exemplary participants of all age groups (now Figure 1—figure supplement 1).

This nicely illustrates that indeed the averaged EEG signal does look like a sleep spindle for both fast sleep spindle types. Further, it shows that for 5- to 6 and 8- to 11-year-olds development-specific and adult-like fast sleep spindles differ in their appearance while they appear very similar in the older age groups. Lastly, by eyeballing the signal during the different sleep spindle types, it also alludes to a link between the adult-like fast sleep spindles and slow oscillations in the 5- to 6- and the 8- to 11-year-olds.

We have now added examples for slow frontal sleep spindles (see Figure 1—figure supplement 1) and revised our figure description and how we refer to this in the main text, e.g.:

Page 5, lines 13–17: “Based on the individual frontal and centro-parietal spindle peak frequencies, we then detected sleep spindle events in frontal and centro-parietal sites which represent the person- and age-specific dominant spindle rhythms per individual (see Figure 1—figure supplement 1 for examples of averaged EEG signals time-locked to the occurrence of these dominant sleep spindles).”

Page 8, lines 8–13: “Therefore, we additionally extracted fast spindles in centro-parietal electrodes by applying fixed frequency criteria between 12.5–16 Hz (henceforth, adult-like fast sleep spindles; see Figure 1—figure supplement 2 for an example raw EEG trace with development-specific and adult-like fast spindles and Figure 1—figure supplement 1 for examples of averaged EEG signals time-locked to development-specific and adult-like fast sleep spindles).”

Similarly, a visualization of the averaged background-corrected power spectra that were used to identify the peak frequencies in frontal and centro-parietal derivations was already depicted in Figure 1A. We have changed the potentially misleading titles (see Figure 1A) and tried to increase the figure size and hope this increases the figure’s informative value.

Lastly, information on sleep spindle activity, time-locked to the slow oscillation down peak is reflected in Figure 3.

We hope we could highlight and clarify where to find the requested descriptive information within the presented data and hope this helps judging the data.

7. Can the authors analyse the phase-amplitude-coupling of the SOs and canonical spindles per age group to extract the phase angle? There seems to be a phase progression from younger to older, which I find quite interesting.

We thank Reviewer 1 for this comment and agree that further specification of the maturational changes in slow oscillation-spindle coupling are an exciting avenue. Theoretically, we could in addition to the time-frequency analyses (TFA) and the peri-event time histograms (PETH) also calculate phase-amplitude coupling measures on our data. However, phase-amplitude coupling measures come with a lot of pitfalls (Aru et al., 2015). In order to allow meaningful physiological interpretation, specifically in the context of age- comparative research, the explanatory power of each component has to be thoroughly established prior to phase-amplitude coupling analyses, e.g., by providing evidence for the existence of rhythmic neural activity at the time point of phase/amplitude extraction, by controlling for the confounding influence of differences in strength and frequency of rhythmic neural events, etc. (see e.g., Hahn et al., 2020; Helfrich et al., 2018). Both our employed measures for slow oscillation-spindle coupling (TFA, PETH) are tailored towards overcoming these drawbacks. The contrast of comparing power during trials with and without slow oscillations controls for power differences between the age groups. The PETHs are constructed after detection of oscillatory events, like sleep spindles and slow oscillation, and thus the caveats that apply to phase and amplitude measures do not affect this method. In addition, with the implemented randomization-test procedure (deviation from randomly shuffled distribution), we are able to assess the temporal reliability and specificity of the results contained in the PETHs. Hence, in our view the employed approaches already provide well-controlled and intuitive results on the temporal interplay between slow oscillations and sleep spindles. Admittedly, while they provide a visualization of the coupling, they do not provide a quantification. However, additional analyses in the form of phase-amplitude coupling would in our view increase the analytic complexity, but not add further information beyond what can be seen with more intuitive and simple approaches like TFAs and PETHs. Since Reviewer 3 also noted that we already report too much data, we would like to stay only with the reported analyses on SO-spindle coupling for this paper.

8. The findings are of course the result of data exploration. This should be clearly acknowledged. It might be helpful to add some robustness analyses to show that the findings are not due to specific choices made during the exploration process.

Thank you for this suggestion. Actually, as stated in the Results section (page 8, lines 13– 16) and the method section (page 50, lines 5–10) not all our analyses were purely based on a data-driven approach, but the main idea that the increased presence of adult-like fast spindles could be associated with the development of slow oscillation-spindle coupling patterns was developed based on the publication including the 5- to 6-year-olds (Joechner et al., 2021). Based on these findings, we decided to extract sleep spindles based on individual peak frequencies and also based on fixed adult-like criteria in the first place. However, we agree that we have not made clear enough in the manuscript which parts were based on ad-hoc data exploration and which were predetermined. We adjusted several lines in the manuscript to make it more transparent when we used a purely data driven approach e.g.,

Page 4, lines 14–16: “Specifically, based on previous analyses (Joechner et al., 2021), we reasoned that the development of fast sleep spindles might be associated with the maturation of SO-spindle coupling.”

Page 21, lines 4–7: “So far, in line with previous observations (Joechner et al., 2021), both analyses on SO-spindle co-occurrence suggested that sleep spindles in the adult-like fast frequency range rather than the more dominant, development-specific fast sleep spindles are coupled to SOs within all age groups.”.

Page 21, lines 10–13: “To capture the “maturational stage” of fast spindles for every participant, we resorted to our observed age-related differences in sleep spindles (Figure 1) and computed the distance between development-specific and adult-like fast sleep spindles within each individual at a given age.”

Regarding the suggestions to add robustness analyses, for the main results that show an association between sleep spindle but not slow oscillation maturity with slow oscillation spindle modulation strength, we have now added an analysis controlling for age (see also comment 5 above) as Supplementary file 1v. Since these results hold and we also find similar results for the two different analyses approaches (partial least squares correlation and the generalized linear mixed-effects models) we are quite confident in this result for our sample.

Further, we also now included a control analysis to check whether our peri-event time histogram results might rely on differing numbers of events as Figure 4—figure supplement 2 (see comment 4). Based on this, and since we already use two different coupling measures that converge in their results and also with the literature (e.g., GarcíaPérez et al., 2022; Hahn et al., 2020), we believe that the slow oscillation-spindle coupling results are robust.

However, as also discussed on page 30, lines 20–24:

“Therefore, we cannot exclude a potentially important contribution of the development and characteristics of SOs for SO-spindle coupling. It may just be that the prominent age-related differences in fast sleep spindles across childhood and adolescence may be relatively more influential and/or that SO features other than the ones examined here may be effective in shaping SO-spindle coupling.”,

we cannot guarantee that our results generalize to other measures and samples. This remains to be tested in the future. As we close in the discussion, page 32, lines 1–4:

“Hence, our findings represent a promising starting point for future research addressing the precise relation between age-related changes in brain structure and function, the emergence of adult-like SO-spindle coupling, and cognitive development across childhood and adolescence.”

9. The time-frequency plots time-locked to SOs do not seem to show a meaningful comparison. This is usually done by comparing two conditions rather than to a baseline. I am not sure we can learn much from them, if there is just a huge cluster across the whole spectrum and time.

Thank you for sharing your concerns about our time frequency analyses. We understand, that seeing the huge clusters might evoke this concern (see also Reviewer 3’s suggestion on decreasing the α level). However, we did not compare our data to a baseline. To reiterate on our approach (see also “5.3.4.1. Time-frequency analyses”, page 52), we indeed did compare different conditions, namely trials with and trials without slow oscillations, on a first level (within participants) to extract maps showing in-/decreases in power during slow oscillation trials. The comparison against zero was done on a second level to establish significance of the t-maps of power differences across participants. Hence, the contrast in the time-frequency plots in Figure 3 (see above page 14) does reflect stable differences in wavelet power between trials with and without slow oscillations, which in our view is a meaningful contrast and has also been used by others (Bastian et al., 2022; Schreiner et al., 2022).

In order to narrow down the cluster, we tried the suggestion of Reviewer 3 and lowered our α level. However, this did not markedly change the size of the cluster (see Author response image 3). There just seems to be robustly higher power during slow oscillation trials in a broad frequency range compared to trials without slow oscillations.

Author response image 3
Results of time frequency analyses with different cluster α.

(A) Cluster α = 0.05 (two-sided, upper panels) and (B) cluster α = 0.01 (two-sided, lower panels). While the cluster shape changed slightly, there was no difference in the overall number of detected clusters.

The spatial, temporal, or frequency extent of clusters by itself is, however, not suited for inferences in the specificity of effects (Sassenhagen and Draschkow, 2019). The cluster size and location typically depend on several factors like the cluster forming procedure, filtering procedures, number of trials etc. Crucially, the exact shape of the cluster is itself typically not subject to statistical testing (Sassenhagen and Draschkow, 2019). Rather a cluster is inherently descriptive and includes with high chance the data points that reflect the true effect ,but at the same time also data points where the null hypothesis is true (Sassenhagen and Draschkow, 2019). Hence, a large cluster does not lessen our descriptive inference that sleep spindle power is enhanced during slow oscillation up peaks. Since the peri-event time histogram analyses converge with the time-frequency analyses we believe, we can be confident about the reported interpretation.

10. For better readability: please do not abbreviate spindles.

Typos: "However, in neither age group, we did" (also: AFAIK "neither" is only used for two alternatives)

Thank you for this suggestion, we have changed this accordingly throughout the manuscript. Further, we have corrected our grammatical error, thank you for pointing this out.

Page 29, lines 10–12: “However, in none of the age groups did we observe strong evidence indicating that sleep spindles outside of the canonical fast spindle frequency range or at other topographic locations couple to distinct phases of the SO.”

Reviewer #2 (Recommendations for the authors):

P3. Line 5: memories are not "transferred" to the cortex but strengthened in the cortex (since already encoded in the cortex).

Thank you for pointing out this error. We have replaced the term “transferred” to “strengthened”.

Page 3 lines 5–9: “Taken together, the complex wave sequence of SO up state and canonical fast sleep spindles, together with hippocampal activity, is considered to provide the scaffold for the precisely timed reactivation of initially fragile hippocampal memory representations and their strengthening in neocortical networks (Diekelmann and Born, 2010; Helfrich et al., 2019; Maingret et al., 2016; Staresina et al., 2015).”

After reading the whole article it is clear why the authors use "canonical" for the "normal" spindles, but it would be easier for readers if the authors the first time when they use the word, explain why and define it.

We thank the reviewer for making us aware that we have not defined explicitly enough what we mean by canonical fast sleep spindles. We agree that this is an important term for the presented research. Therefore, we already attempted to define canonical fast and slow spindles in the previous manuscript version by naming their frequency range and the topographical predominance in parentheses. In order to make clearer that we mean the fast spindle range that has been observed for young adults, we now updated the revised manuscript where appropriate, e.g.:

Page 2, lines 13–18: “Far from being an epiphenomenon, accumulating evidence suggests that the synchronization of canonical fast sleep spindles (i.e., spindles defined in young adults with a frequency of ≈ 12.5–16 Hz and a centro-parietal predominance; Cox et al., 2017), precisely during the up state of SOs provides an essential mechanism for neural communication, e.g., supporting systems memory consolidation during sleep (Hahn et al., 2020; Helfrich et al., 2018; Latchoumane et al., 2017; Mölle et al., 2002; Muehlroth et al., 2019)”

Page 2, line 23–page 3, line 3: “In addition to canonical fast sleep spindles, there is substantial evidence for a canonical slow sleep spindle type (i.e., spindles defined in young adults with a frequency of ≈ 9–12.5 Hz and a frontal predominance; Cox et al., 2017) in the human surface electroencephalogram (EEG; De Gennaro and Ferrara, 2003; Fernandez and Lüthi, 2020).”

Page 6, line 18–page 7, line 8: “Crucially, despite being faster, the frequency of the fast centroparietal spindles was specific to the age of the participants in a way that at younger ages, these dominant fast sleep spindles were not yet in the range of canonical fast spindles (i.e., as on average observed in adults with a frequency of ≈ 12.5–16 Hz; Cox et al., 2017; see Figure 1 A–B). For the vast majority of 14- to 18- and 20- to 26-year-olds, individually identified, dominant centro-parietal sleep spindles indeed matched the canonical fast sleep spindle frequency range (≈ 12.5–16 Hz, see Figure 1 A–B). For the majority of children though, dominant fast centro-parietal spindles were nested in the canonical slow spindle range (i.e., as on average observed in adults with a frequency of ≈ 9–12.5 Hz; Cox et a., 2017) and thus manifested in a development-specific fashion. Hence, the term “development-specific” will be employed to refer to individually determined, dominant fast centro-parietal sleep spindles across all age groups in the following. In contrast, since we also found dominant fast centro-parietal sleep spindles between 12.5–16 Hz in our adult sample, we will denote sleep spindle activity in this canonical fast spindle range and events detected exclusively within this canonical frequency range in centro-parietal sites as “adult-like” in our sample across all age groups.”

Page 8, lines 5–13: “After having characterized prevailing , development-specific fast centroparietal spindles across all our age groups, we were interested in how they differed from canonical fast spindles, i.e., those commonly, and also here, found in young adults with a frequency of ≈12.5–16 Hz (see Figure 1 for the present adult sample, cf. Cox et al., 2017; Ujma et al., 2015). Therefore, we additionally extracted fast spindles in centro-parietal electrodes by applying fixed frequency criteria between 12.5–16 Hz (henceforth, adult-like fast sleep spindles; see Figure 1—figure supplement 2 for an example raw EEG trace with development specific and adult-like fast spindles and Figure 1—figure supplement 1 for examples of averaged EEG signals time-locked to development-specific and adult-like fast sleep spindles).”

We hope that this helps improving clarity.

Reviewer #3 (Recommendations for the authors):

Title: The title is rather unspecific with respect the age range and could be specified further.

Thank you for this suggestion, we have adapted the title accordingly to: “Sleep spindle maturity promotes slow oscillation-spindle coupling across child and adolescent development”

Figures: Almost all plots showing individual data points, means and raincloud plots are very hard to interpret and could need a small overhaul. In particular Figure 2, does not allow a visual assessment of the differences between cortical areas.

We agree that our plots were overcrowded with information and that too many plots were squeezed into too little space making it hard to interpret them. In the course of the revision, we have tried to reduce the amount of information within individual plots and to re-arrange them. Further, we rearranged the factors that are plotted in Figure 2 and hope that this improves the readability of the presented figures and increases their informative power.

Page 1, line 4: For consistency, the authors could also add the frequency range of SOs here.

Thank you for this suggestion, we have added this information in the abstract.

Page 1, lines 2–4: “The synchronization of canonical fast sleep spindle activity (12.5–16 Hz, adult-like) precisely during the slow oscillation (0.5–1 Hz) up peak is considered an essential feature of adult non-rapid eye movement sleep.”

Page 2, line 11: Here it is stated that how the coupling develops is an open question. However, at the end of the next paragraph, a previous publication is reported that shows how the coupling improves across childhood. Is this open question still valid then?

Moreover, throughout the introduction, the authors repeat this open question several times with slights variations in the phrasing. I wonder if this is necessary or if the open question can be asked centrally before the final paragraph of the introduction.

We agree, that we have stated the same broad open question very extensively in the introduction. We have revised this and rephrased the question in the first (page 2, lines 1112), second (page 3, lines 1013), and third paragraph (page 4, lines 911) to:

1) “Yet, little is known about how this precise coalescence develops across childhood and adolescence.”

2) “However, the precise coupling of sleep spindle activity to SOs, described above, does not seem to be fully present and functional from early childhood on. Recent evidence in rodents and humans indicates that the temporal co-ordination of SOs and spindles improves across childhood and adolescence (García-Pérez et al., 2022; Hahn et al., 2020; Joechner et al., 2021).”

3) “Nevertheless, it is still unclear how developments in sleep spindles and SOs interact to promote precise, adult-like temporal synchronization of sleep spindles during SOs across childhood and adolescence.”

Even though we have rephrased the original question in the second paragraph (page 3, lines 1013) to a declarative sentence (see (2)) and the existence of some excellent literature, we are convinced, that the very broad question on how slow oscillation-spindle coupling develops is still valid. We are at the moment aware of only a handful of papers that examine slow oscillation-spindle coupling in a developmental context (e.g., García-Pérez et al., 2022; Hahn et al., 2020, 2022; Joechner et al., 2021; Kurz et al., 2021; Piantoni et al., 2013). Amongst these, most publications focus on late childhood and adolescence. While the mentioned research provides important insights into slow oscillation-spindle coupling development, a lot of open, more specific, questions remain e.g., how is coupling expressed in young childhood and infancy? What drives developments in slow oscillation coupling across maturation? How is this connected to memory consolidation?

We hope that our adjustments removed redundancies and improve introducing our narrow question on how sleep spindle and slow oscillation developments relate to developments in their coupling (see (3) and page 4, lines 911).

Page 6, line 4: Please see my comment in the public review. I think the assessment of average frequency of spindle events is redundant or even circular as they are based on the power spectra.

We thank Reviewer 3 for highlighting redundancies in our reported results. We agree that reporting the results of both the peak frequencies and the average frequencies of the individually identified sleep spindles is unnecessary.

However, we have several reasons for assessing the average frequency of sleep spindle events:

First, we frequently encounter the problem that detection based on adult-like criteria can lead to the identification of sleep spindles below the lower boundary of this window; specifically, for young children whose peak frequencies are markedly lower. This leads to the phenomenon that average spindle frequencies can be lower (theoretically also higher) than the upper and lower boundaries of the intended search frequency window if these powerful events are included and dominate. Hence, assessing the average frequency of adult-like fast sleep spindles is in our view important to validate our methodical approach and to ensure that we detected sleep spindles in the range of interest in all age groups.

Second, we consider the comparison of the frequencies of spindles detected based on individual peak frequencies and fixed adult-like criteria central to quantify sleep spindle development. Given the absence of spectral peaks for adult-like spindles, the average frequencies of detected events for both sleep spindle types represents a comparable measure in both spindle types.

Therefore, we have now decided to concentrate on reporting the average frequencies of spindle events in the main text and to only report the statistical results from the peak frequency comparisons in Supplementary files 1a and b for completeness.

Page 6, line 18: The introduction of development-specific feels unnecessary and to some degree arbitrary. At least for me, it only led to overly complicated/cluttered sentences. Is this term really required or can it be omitted while the main message is conveyed properly.

Thank you for sharing your concerns with the introduced terminology. We agree that in our attempt to be maximally descriptive and discriminative in our terminology some long terms have emerged. However, in our view the term “development-specific” conveys a central message: despite between-person differences, dominant fast centro-parietal sleep spindle activity manifest very differently across (lifespan) development (cf. Cox et al., 2017; Muehlroth and Werkle‐Bergner, 2020). Hence, on a more general scale, this term does not only apply to the phenomenon that we find slower fast sleep spindles in young children in the context of this paper but is also applicable to different time scales and contexts and hopefully fosters usage of adaptive methods to capture a variable phenomenon.

We consider the contrast of dominant sleep spindles being manifested in a development specific fashion across maturation and yet a stable presence of slow oscillation-spindle coupling for spindles in the adult-like fast spindle range one of the main conundrums. Hence, to highlight the fact of a low presence of coupled adult-like fast sleep spindles in the face of distinctly manifested dominant sleep spindles, we would like to stick to the term “development-specific” in the context of this paper in cases where we directly contrast results based on the different sleep spindle detection approaches. We believe that using these descriptive and discriminative terms helps the reader to better follow the analysis steps.

However, we agree that these terms are very long. Therefore, we went through the manuscript in the course of this revision and tried to reduce its usage in all sections except the Results section. We hope this makes sentences in these parts of the manuscript less cluttered.

Page 10, Figure 1: It would help the reader if the age ranges are illustrated in subpanel A. Moreover, as stated above, symbols for the different spindle types and the marginal means are very hard to distinguish. Also, I was at first not aware that the adult-like spindles even had a rain cloud. Perhaps the authors can rearrange/resize the figure.

Related to subpanel A, its known that adults sometimes do not show slow spindle peaks or shows both a slow and fast spindle peak at one electrode (e.g., Mölle et al., 2011). At least the former seems to be the same here. How did the authors proceed if that was the case? I was wondering if it would be more informative to sort the individual subjects peak frequency per age group.

We thank Reviewer 3 for his suggestions to improve our figures, we have tried to implement them and hope this makes them easier to read and more informative (see comment on page 24 above). Regarding the suggestion for Figure 1A (see page 25 of this response letter for the revised Figure 1), we agree that it is most informative to sort participants by age groups. Indeed, this is what we already did. As indicated by in the figure text, page 12, line 4: “Data are ordered by age from bottom to top (y-axis)”, we did sort the power spectral plots not only by age group but also within age group by age (in months).

Regarding the non-trivial question on the issue of missing or several peaks in the sleep spindle frequency range for the identification of slow and fast sleep spindles, we consider this one of the core problems when differentiating slow and fast spindles based on predefined frequency criteria or across averaged topographies (Cox et al., 2017; Joechner et al., 2021). This is why we have neither identified peaks nor spindle events as slow or fast based on their absolute frequency but we took a slightly different approach that we want to reiterate and elaborate shortly in the following.

As also described on pages 50 /51 under “5.3.3 Sleep spindle and slow oscillation detection”, we implemented the identification for spectral peaks between 9 and 16 Hz separately for averaged frontal and centro-parietal electrodes, where usually slow and fast sleep spindles dominate, respectively, and based on the background-corrected power spectra. We then identified the maximum peak between 9 and 16 Hz, that also had a zero crossing in the 1st derivative, as the dominant peak frequency. Hence, in case of the presence of a less prominent second (third…) peak, this peak (s) was/were not informing the peak frequency measure. However, since we used the detected frontal and centro-parietal spectral peaks as center frequencies of a quite broad frequency window (± 1.5 Hz) for spindle event detection in frontal or centro-parietal electrodes, spindles at slightly different frequencies than the peak frequency were also detected.

Importantly, in every participant, we found one prominent peak in frontal and centro-parietal background corrected power spectra. We controlled peak detection by our algorithm through visual inspection of all power spectra and in all cases the algorithm captured the prominent peak per topography and individual. Further, only a rare number of participants showed a second, less prominent and flatter peak in averaged frontal or centro-parietal power spectra, that was in close proximity to the more prominent peak.

Crucially, at this stage, we do not yet classify peaks or spindles detected based on the peak frequencies as slow or fast based on their frequency (so below or above a certain frequency threshold) as, to the best of our knowledge, done in Mölle et al. (2011), or on topography alone. Therefore, we did not run into the problem of missing slow or fast peaks.

Rather, based on the assumption that slow(er) spindles usually dominate in frontal and fast(er) spindles in centro-parietal areas, we then tested whether spectral peak frequencies and frontal spindles detected based on the peak frequency would have a slower frequency than spectral peaks and centro-parietal spindles, that were extracted on the individual peak frequencies. Since this was statistically evident, the results imply to us that, based on both topography and frequency (Anderer et al., 2001; Cox et al., 2017; Gibbs and Gibbs, 1950; Mölle et al., 2011), we find evidence for slow frontal and fast-centro-parietal sleep spindles. This does not automatically mean that either frontal or centro-parietal spindles are in the canonical slow (912.5 Hz) or fast spindle range (12.516 Hz Hz) for every participant. While we explicitly focus on this feature for centro-parietal spindles in the context of maturation, the inverse phenomenon (a slower frontal spindle peak that is faster than the canonical slow spindle peak) could also apply.

Taken together, by detecting peaks in background corrected power spectra separately for frontal and centro-parietal electrodes, we increased the chance of detecting prominent peaks that reflect true rhythmic neural activity. While our results imply that these peaks and the derived sleep spindles were on average slower in frontal than centro-parietal derivations, this does not mean that either these slow frontal or centro-parietal peaks or sleep spindle events would be considered as slow or fast based on canonical definitions.

Page 14, line 4: What do the authors mean by "more precise"? The size of the cluster or its location with respect to the SO up-state? Is it really possible to make conclusion about the precision using a TFR-based analysis affected by temporal and frequency smearing?

In general, we mean with “more precise” the temporal clustering closer to the slow oscillation up peak on a descriptive level (time-frequency and peri-event time histogram (PETH) analyses). We are aware that clusters from cluster-based permutation tests do not justify statistical inferences about the spatio-temporal extent or location of clusters, while on a descriptive level, description of effect shapes and locations is justified (Sassenhagen and Draschkow, 2019). Further, in the case of our time-frequency analyses, just like in every time-frequency analysis, temporal and frequency smearing limit exact interpretability of frequency and temporal boundaries (Cohen, 2019). We have chosen a 12-cycle wavelet which rather emphasizes good frequency resolution over temporal precision. However, while this limits identification of exact time and frequency differences of effects, it allows careful description of the effects. Therefore, we already have tried to be careful in our description of the time-frequency analyses by using terms as “seemed stronger and more precise” (actually now: " seemed to be stronger and more precise during the SO up peak”) on page 16, line 16 and highlighting that we are on a “descriptive level” (page 16 line 12). The same applies to the description of the PETH results, where the concerns regarding temporal or frequency smearing do not apply. Since both analyses converge in terms of descriptive interpretation of the temporal clustering of increased and decreased sleep spindle occurrence we feel confident in describing our results in terms of precision.

In order to resolve issues with our use of the term “more precise” and to avoid misleading the readers, we revised the respective parts in the manuscript to be more explicit, e.g.:

Page 26, lines 13–16: “Surprisingly, the inspection of SO-spindle coupling patterns suggested synchronization being driven by spindles in the adult-like, canonical fast spindle range—even in the younger age groups but notably less precisely timed during the SO cycle.”

Page 27, lines12/13: “Overall, our findings suggest that the temporally precise synchronization of sleep spindles during the up peak of SOs might not be an inherent feature of the thalamocortical system.”

Page 29, lines 16–20: “Surprisingly though, the modulation of adult-like fast sleep spindles during SOs was well detectable across all age groups with this co-occurrence pattern becoming more precisely timed with SO up and down peaks with older age—despite huge age-related differences in their occurrence rate and the fact that adult-like fast sleep spindles showed very low density in both child cohorts (see Figure 1).” Page 14, Figure 2.

A comparison between SO vs. SO-free intervals includes a strong difference in overall spectral power, hence the wide significant cluster. The authors could consider decreasing their α-value further to tease out clusters more clearly. Furthermore, the dashed window highlighting the spindle range is not really visible.

Page 14, Figure 2. A comparison between SO vs. SO-free intervals includes a strong difference in overall spectral power, hence the wide significant cluster. The authors could consider decreasing their α-value further to tease out clusters more clearly. Furthermore, the dashed window highlighting the spindle range is not really visible.

Thank you for this suggestion. We have tried decreasing the α level for the cluster-based permutation test on the between-participant level, however, the cluster sizes did not change considerably (Author response image 3). Therefore, we decided to stay with the original α.

Concerning the figures on our time-frequency results, we have now intensified the color of the window indicating the spindle frequency range and hope this improves our figures (see Figure 3).

Page 16, Figure 4: Again, this is reflecting my personal opinion, but this the central result and it somehow feels like it does not get enough attention. Anyhow, what does the horizontal jiggly line represent? It is also stated that the empirical PETHs are compared to a surrogate distribution, however, is that really necessary? The authors have SO-free intervals that could be used for a cluster-based permutation test as well.

The horizontal jiggly line in the peri-event time histograms (PETH) reflects the surrogate distribution that was the comparison condition to establish statistical significance. We describe the line in the figure legend with the following sentence:

“Green asterisks mark increased spindle occurrence (positive cluster, cluster-based permutation test, cluster α <.05, two-sided test) and red asterisks mark decreased spindle occurrence (negative cluster, cluster-based permutation test, cluster α <.05, two-sided test) compared to random occurrence (horizontal line)”.

We now added more explicitly that the horizontal line represents the surrogate comparison distribution and hope this increases interpretation:

Page 19, lines 4–7: “Green asterisks mark increased spindle occurrence (positive cluster, cluster-based permutation test, cluster α <.05, two-sided test) and red asterisks mark decreased spindle occurrence (negative cluster, cluster-based permutation test, cluster α < .05, two-sided test) compared to a surrogate distribution representing random occurrence (horizontal line).”

Concerning the choice of reference distribution to statistically determine whether spindle occurrence during slow oscillation windows was temporally clustered, we consider randomly shuffling the original data a straight forward, easy procedure to obtain a comparison distribution that misses temporal structure but otherwise contains the same properties as the original distribution (Lancaster et al., 2018; for similar approaches, see Bastian et al., 2022; Muehlroth et al., 2019). Therefore, it allows to easily determine whether there is a temporal structure in the data. We agree, that alternatively, we could have also employed slow oscillation-free trials to construct comparison distributions that would be suitable to test whether a temporal structure in the PETHs would be specific to slow oscillation trials.

Lastly, since we agree that these are important results, we have changed the headings on page 16 to:

“2.3 Temporal modulation of adult-like fast sleep spindle power during slow oscillations is present across all ages” and on page 17 to “2.4 At older age, development specific fast sleep spindle occurrence is more strongly modulated by slow oscillations while slow oscillation-adult-like fast spindle coupling becomes temporally more precise”.

We hope this helps highlighting the results.

Page 19, Figure 6: Can the author please explain what the main message of subpanel A is? This somehow eludes me. Also, how come there is not separation into the different age groups?

Thank you for making us aware of the lack of clarity of our description and plots on this part of the partial least squares correlation (PLSC). As described under “5.4.2. Association between sleep spindle and slow oscillation maturity with slow oscillation-spindle coupling”, page 55/56, we determined the reliability of the sleep spindle maturity profile, that was associated with a specific pattern of power modulations during slow oscillations by means of:

“the correlation between the latent time-frequency variable and the three raw spindle maturity variables. These values are comparable to the spindle maturity weights and indicate whether the association between the time-frequency pattern and individual spindle maturity variables is in the same or different direction (McIntosh and Lobaugh, 2004). Stability of these correlation patterns was determined by their bootstrap estimated 95% confidence intervals (McIntosh and Lobaugh, 2004).”

Figure 6A displays the pattern of how the individual fast sleep spindle maturity features are related to the corresponding coupling pattern (Figure 6B). In this case all features were reliably and in the same way related to the slow oscillation-spindle coupling pattern. We have tried to improve the clarity of Figure 6 by revising its explanation in the main text.

Further, there is no separation by age group since we considered all age groups within one PLS analysis as noted on page 55 under “5.4.2. Association between sleep spindle and slow oscillation maturity with slow oscillation-spindle coupling”. Hence Figure 6A and B reflect the associated pattern of sleep spindle maturity and slow oscillation-spindle coupling across all age groups. We are aware that this means that there are participants with differing dependencies in the analyses. However, this only acts against us in terms of power since the participants from the longitudinal cohort are treated as independent individuals. While we had this information described in the method section, we admittedly did not mention this clearly in the Results section. We have now added this information also in the Results section on page 23, lines 11-24:

“For one, we examined the relation between the pattern of power modulations in the spindle frequency range (9–16 Hz) during the complete SO trial (down peak ± 1.2 sec, cf. time frequency t-maps, Figure 3) with the fast spindle maturity scores for sleep spindle frequency, density, and amplitude using a partial least squares correlation (PLSC; Krishnan et al., 2011) across all participants. This analysis provides pairs of SO-spindle coupling profiles and associated, multivariate patterns of spindle maturity scores (spindle maturity profile). Based on a permutation test, we identified one significant pair of a spindle maturity profile (Figure 6A) and a specific time-frequency SO-spindle coupling pattern (Figure 6B; singular vector pair p <.001). All fast spindle maturity measures contributed reliably and positively to this significant, positive correlation (as indicated by the direction of values in the spindle maturity profile and the non-zero crossing confidence intervals in Figure 6A). As can be inferred from Figure 6B, the SO-spindle coupling pattern suggests, that higher fast sleep spindle maturity in all features (Figure 6A) was associated with a more adult-like SO-spindle coupling pattern, reflected in: […]”

References

Anderer, P., Klösch, G., Gruber, G., Trenker, E., Pascual-Marqui, R. D., Zeitlhofer, J., Barbanoj, M. J., Rappelsberger, P., and Saletu, B. (2001). Low-resolution brain electromagnetic tomography revealed simultaneously active frontal and parietal sleep spindle sources in the human cortex. Neuroscience, 103(3), 581–592. https://doi.org/10.1016/S0306-4522(01)00028-8

Aru, J., Aru, J., Priesemann, V., Wibral, M., Lana, L., Pipa, G., Singer, W., and Vicente, R. (2015). Untangling cross-frequency coupling in neuroscience. Current Opinion in Neurobiology, 31, 51–61. https://doi.org/10.1016/j.conb.2014.08.002

Attarian, H., and Viola-Saltzman, M. (Hrsg.). (2020). Sleep Disorders in Women: A Guide to Practical Management. Springer International Publishing. https://doi.org/10.1007/978-3-030-40842-8

Barakat, M., Doyon, J., Debas, K., Vandewalle, G., Morin, A., Poirier, G., Martin, N., Lafortune, M., Karni, A., Ungerleider, L. G., Benali, H., and Carrier, J. (2011). Fast and slow spindle involvement in the consolidation of a new motor sequence. Behavioural Brain Research, 217(1), 117–121. https://doi.org/10.1016/j.bbr.2010.10.019

Bastian, L., Samanta, A., Ribeiro de Paula, D., Weber, F. D., Schoenfeld, R., Dresler, M., and Genzel, L. (2022). Spindle–slow oscillation coupling correlates with memory performance and connectivity changes in a hippocampal network after sleep. Human Brain Mapping, 43(13), 3923–3943. https://doi.org/10.1002/hbm.25893

Bates, D., Mächler, M., Bolker, B., and Walker, S. (2015). Fitting linear mixed-effects models using lme4. Journal of Statistical Software, 67(1), 1–48. https://doi.org/10.18637/jss.v067.i01

Campbell, I. G., and Feinberg, I. (2016). Maturational patterns of σ frequency power across childhood and adolescence: A longitudinal study. Sleep, 39(1), 193–201. https://doi.org/10.5665/sleep.5346

Cohen, M. X. (2019). A better way to define and describe Morlet wavelets for time-frequency analysis. NeuroImage, 199, 81–86. https://doi.org/10.1016/j.neuroimage.2019.05.048

Cowan, E., Liu, A., Henin, S., Kothare, S., Devinsky, O., and Davachi, L. (2020). Sleep spindles promote the restructuring of memory representations in ventromedial prefrontal cortex through enhanced hippocampal-cortical functional connectivity. Journal of Neuroscience, 40(9), 1909–1919. https://doi.org/10.1523/JNEUROSCI.1946-19.2020

Cox, R., Schapiro, A. C., Manoach, D. S., and Stickgold, R. (2017). Individual differences in frequency and topography of slow and fast sleep spindles. Frontiers in Human Neuroscience, 11, Article 433. https://doi.org/10.3389/fnhum.2017.00433

D’Atri, A., Novelli, L., Ferrara, M., Bruni, O., and De Gennaro, L. (2018). Different maturational changes of fast and slow sleep spindles in the first four years of life. Sleep Medicine, 42, 73–82. https://doi.org/10.1016/j.sleep.2017.11.1138

De Gennaro, L., and Ferrara, M. (2003). Sleep spindles: An overview. Sleep Medicine Reviews, 7(5), 423–440. https://doi.org/10.1053/smrv.2002.0252

Diekelmann, S., and Born, J. (2010). The memory function of sleep. Nature Reviews Neuroscience, 11(2), 114–126. https://doi.org/10.1038/nrn2762

Fernandez, L. M. J., and Lüthi, A. (2020). Sleep spindles: Mechanisms and functions. Physiological Reviews, 100(2), 805–868. https://doi.org/10.1152/physrev.00042.2018

Franco, P., Putois, B., Guyon, A., Raoux, A., Papadopoulou, M., Guignard-Perret, A., Bat-Pitault, F., Hartley, S., and Plancoulaine, S. (2020). Sleep during development: Sex and gender differences. Sleep Medicine Reviews, 51, 101276. https://doi.org/10.1016/j.smrv.2020.101276

García-Pérez, M. A., Irani, M., Tiznado, V., Bustamante, T., Inostroza, M., Maldonado, P. E., and Valdés, J. L. (2022). Cortico-Hippocampal Oscillations Are Associated With the Developmental Onset of Hippocampal-Dependent Memory. Frontiers in Neuroscience, 16, 891523. https://doi.org/10.3389/fnins.2022.891523

Gibbs F.A., and Gibbs E.L. (1950). Atlas of Electroencephalography. Reading, MA.: Addison‐Wesley, 1950.

Hahn, M. A., Bothe, K., Heib, D., Schabus, M., Helfrich, R. F., and Hoedlmoser, K. (2022). Slow oscillation– spindle coupling strength predicts real-life gross-motor learning in adolescents and adults. eLife, 11, Article e66761. https://doi.org/10.7554/eLife.66761

Hahn, M. A., Heib, D., Schabus, M., Hoedlmoser, K., and Helfrich, R. F. (2020). Slow oscillation-spindle coupling predicts enhanced memory formation from childhood to adolescence. eLife, 9, Article e53730. https://doi.org/10.7554/eLife.53730

Hahn, M. A., Joechner, A.-K., Roell, J., Schabus, M., Heib, D. P., Gruber, G., Peigneux, P., and Hoedlmoser, K. (2019). Developmental changes of sleep spindles and their impact on sleep‐dependent memory consolidation and general cognitive abilities: A longitudinal approach. Developmental Science, 22(1), Article e12706. https://doi.org/10.1111/desc.12706

Helfrich, R. F., Lendner, J. D., Mander, B. A., Guillen, H., Paff, M., Mnatsakanyan, L., Vadera, S., Walker, M. P., Lin, J. J., and Knight, R. T. (2019). Bidirectional prefrontal-hippocampal dynamics organize information transfer during sleep in humans. Nature Communications, 10(1), Article 3572. https://doi.org/10.1038/s41467-019-11444-x

Helfrich, R. F., Mander, B. A., Jagust, W. J., Knight, R. T., and Walker, M. P. (2018). Old brains come uncoupled in sleep: Slow wave-spindle synchrony, brain atrophy, and forgetting. Neuron, 97(1), 221-230.e4. https://doi.org/10.1016/j.neuron.2017.11.020

Joechner, A.-K., Wehmeier, S., and Werkle-Bergner, M. (2021). Electrophysiological indicators of sleepassociated memory consolidation in 5‐ to 6-year-old children. Psychophysiology, 58(8), Article e13829. https://doi.org/10.1111/psyp.13829

Krishnan, A., Williams, L. J., McIntosh, A. R., and Abdi, H. (2011). Partial Least Squares (PLS) methods for neuroimaging: A tutorial and review. NeuroImage, 56(2), 455–475. https://doi.org/10.1016/j.neuroimage.2010.07.034

Kurz, E.-M., Conzelmann, A., Barth, G. M., Renner, T. J., Zinke, K., and Born, J. (2021). How do children with autism spectrum disorder form gist memory during sleep? A study of slow oscillation–spindle coupling. Sleep, 44(6), Article zsaa290. https://doi.org/10.1093/sleep/zsaa290

Kuznetsova, A., Brockhoff, P. B., and Christensen, R. H. B. (2017). lmerTest Package: Tests in linear mixed effects models. Journal of Statistical Software, 82(13). https://doi.org/10.18637/jss.v082.i13

Lancaster, G., Iatsenko, D., Pidde, A., Ticcinelli, V., and Stefanovska, A. (2018). Surrogate data for hypothesis testing of physical systems. Physics Reports, 748, 1–60. https://doi.org/10.1016/j.physrep.2018.06.001

Latchoumane, C.-F. V., Ngo, H.-V. V., Born, J., and Shin, H.-S. (2017). Thalamic spindles promote memory formation during sleep through triple phase-locking of cortical, thalamic, and hippocampal rhythms. Neuron, 95(2), 424-435.e6. https://doi.org/10.1016/j.neuron.2017.06.025

Lindenberger, U., and Pötter, U. (1998). The complex nature of unique and shared effects in hierarchical linear regression: Implications for developmental psychology. Psychological Methods, 3(2), 218– 230.

Lindenberger, U., von Oertzen, T., Ghisletta, P., and Hertzog, C. (2011). Cross-sectional age variance extraction: What’s change got to do with it? Psychology and Aging, 26(1), 34–47. https://doi.org/10.1037/a0020525

Maingret, N., Girardeau, G., Todorova, R., Goutierre, M., and Zugaro, M. (2016). Hippocampo-cortical coupling mediates memory consolidation during sleep. Nature Neuroscience, 19(7), 959–964. https://doi.org/10.1038/nn.4304

McIntosh, A. R., and Lobaugh, N. J. (2004). Partial least squares analysis of neuroimaging data: Applications and advances. NeuroImage, 23, S250–S263. https://doi.org/10.1016/j.neuroimage.2004.07.020

Mölle, M., Bergmann, T. O., Marshall, L., and Born, J. (2011). Fast and slow spindles during the sleep slow oscillation: Disparate coalescence and engagement in memory processing. Sleep, 34(10), 1411– 1421. https://doi.org/10.5665/SLEEP.1290

Mölle, M., Marshall, L., Gais, S., and Born, J. (2002). Grouping of spindle activity during slow oscillations in human non-rapid eye movement sleep. The Journal of Neuroscience, 22(24), 10941–10947. https://doi.org/10.1523/JNEUROSCI.22-24-10941.2002

Mong, J. A., and Cusmano, D. M. (2016). Sex differences in sleep: Impact of biological sex and sex steroids. Philosophical Transactions of the Royal Society B: Biological Sciences, 371(1688), 20150110. https://doi.org/10.1098/rstb.2015.0110

Muehlroth, B. E., Sander, M. C., Fandakova, Y., Grandy, T. H., Rasch, B., Shing, Y. L., and Werkle-Bergner, M. (2019). Precise slow oscillation–spindle coupling promotes memory consolidation in younger and older adults. Scientific Reports, 9(1), Article 1940. https://doi.org/10.1038/s41598-01836557-z

Muehlroth, B. E., and Werkle‐Bergner, M. (2020). Understanding the interplay of sleep and aging: Methodological challenges. Psychophysiology, 57(3), Article e13523. https://doi.org/10.1111/psyp.13523

Nicolas, A., Petit, D., Rompré, S., and Montplaisir, J. (2001). Sleep spindle characteristics in healthy subjects of different age groups. Clinical Neurophysiology, 112(3), 521–527. https://doi.org/10.1016/S1388-2457(00)00556-3

Niknazar, M., Krishnan, G. P., Bazhenov, M., and Mednick, S. C. (2015). Coupling of thalamocortical sleep oscillations are important for memory consolidation in humans. PLoS ONE, 10(12), e0144720. https://doi.org/10.1371/journal.pone.0144720

Olbrich, E., Rusterholz, T., LeBourgeois, M. K., and Achermann, P. (2017). Developmental changes in sleep oscillations during early childhood. Neural Plasticity, 2017, Article ID 6160959. https://doi.org/10.1155/2017/6160959

Peiffer, A., Brichet, M., De Tiège, X., Peigneux, P., and Urbain, C. (2020). The power of children’s sleep: Improved declarative memory consolidation in children compared with adults. Scientific Reports,

10(1), Article 9979. https://doi.org/10.1038/s41598-020-66880-3

Piantoni, G., Astill, R. G., Raymann, R. J. E. M., Vis, J. C., Coppens, J. E., and Van Someren, E. J. W. (2013). Modulation of γ and spindle-range power by slow oscillations in scalp sleep EEG of children. International Journal of Psychophysiology, 89(2), 252–258. https://doi.org/10.1016/j.ijpsycho.2013.01.017

Purcell, S. M., Manoach, D. S., Demanuele, C., Cade, B. E., Mariani, S., Cox, R., Panagiotaropoulou, G., Saxena, R., Pan, J. Q., Smoller, J. W., Redline, S., and Stickgold, R. (2017). Characterizing sleep spindles in 11,630 individuals from the National Sleep Research Resource. Nature Communications, 8(1), Article 15930. https://doi.org/10.1038/ncomms15930

Rasch, B., and Born, J. (2013). About sleep’s role in memory. Physiological Reviews, 93(2), 681–766. https://doi.org/10.1152/physrev.00032.2012

Sassenhagen, J., and Draschkow, D. (2019). Cluster‐based permutation tests of MEG/EEG data do not establish significance of effect latency or location. Psychophysiology, 56(6), e13335. https://doi.org/10.1111/psyp.13335

Scholle, S., Zwacka, G., and Scholle, H. C. (2007). Sleep spindle evolution from infancy to adolescence. Clinical Neurophysiology, 118(7), 1525–1531. https://doi.org/10.1016/j.clinph.2007.03.007

Schreiner, T., Kaufmann, E., Noachtar, S., Mehrkens, J.-H., and Staudigl, T. (2022). The human thalamus orchestrates neocortical oscillations during NREM sleep. Nature Communications, 13(1), 5231. https://doi.org/10.1038/s41467-022-32840-w

Shinomiya, S., Nagata, K., Takahashi, K., and Masumura, T. (1999). Development of sleep spindles in young children and adolescents. Clinical Electroencephalography, 30(2), 39–43. https://doi.org/10.1177/155005949903000203

Staresina, B. P., Bergmann, T. O., Bonnefond, M., van der Meij, R., Jensen, O., Deuker, L., Elger, C. E., Axmacher, N., and Fell, J. (2015). Hierarchical nesting of slow oscillations, spindles and ripples in the human hippocampus during sleep. Nature Neuroscience, 18(11), 1679–1686. https://doi.org/10.1038/nn.4119

Ujma, P. P., Gombos, F., Genzel, L., Konrad, B. N., Simor, P., Steiger, A., Dresler, M., and Bódizs, R. (2015). A comparison of two sleep spindle detection methods based on all night averages: Individually adjusted vs. fixed frequencies. Frontiers in Human Neuroscience, 9, Article 52. https://doi.org/10.3389/fnhum.2015.00052

Urbain, C., De Tiège, X., Op De Beeck, M., Bourguignon, M., Wens, V., Verheulpen, D., Van Bogaert, P., and Peigneux, P. (2016). Sleep in children triggers rapid reorganization of memory-related brain processes. NeuroImage, 134, 213–222. https://doi.org/10.1016/j.neuroimage.2016.03.055

Wilhelm, I., Prehn-Kristensen, A., and Born, J. (2012). Sleep-dependent memory consolidation – What can be learnt from children? Neuroscience and Biobehavioral Reviews, 36(7), 1718–1728. https://doi.org/10.1016/j.neubiorev.2012.03.002

Wilhelm, I., Rose, M., Imhof, K. I., Rasch, B., Büchel, C., and Born, J. (2013). The sleeping child outplays the adult’s capacity to convert implicit into explicit knowledge. Nature Neuroscience, 16(4), 391–393. https://doi.org/10.1038/nn.3343

Zhang, Z. Y., Campbell, I. G., Dhayagude, P., Espino, H. C., and Feinberg, I. (2021). Longitudinal analysis of sleep spindle maturation from childhood through late adolescence. Journal of Neuroscience, 41(19), 4253–4261. https://doi.org/10.1523/JNEUROSCI.2370-20.2021

https://doi.org/10.7554/eLife.83565.sa2

Article and author information

Author details

  1. Ann-Kathrin Joechner

    Center for Lifespan Psychology, Max Planck Institute for Human Development, Berlin, Germany
    Contribution
    Conceptualization, Data curation, Software, Formal analysis, Methodology, Writing – original draft, Project administration, Writing – review and editing
    For correspondence
    joechner@mpib-berlin.mpg.de
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0003-4962-1089
  2. Michael A Hahn

    1. Department of Psychology, Laboratory for Sleep, Cognition and Consciousness Research, University of Salzburg, Salzburg, Austria
    2. Centre for Cognitive Neuroscience Salzburg (CCNS), University of Salzburg, Salzburg, Austria
    3. Hertie-Institute for Clinical Brain Research, University Medical Center Tuebingen, Tuebingen, Germany
    Contribution
    Conceptualization, Data curation, Writing – review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-3022-0552
  3. Georg Gruber

    1. Department of Psychiatry and Psychotherapy, Medical University of Vienna, Vienna, Austria
    2. The Siesta Group, Vienna, Austria
    Contribution
    Software, Writing – review and editing
    Competing interests
    No competing interests declared
  4. Kerstin Hoedlmoser

    1. Department of Psychology, Laboratory for Sleep, Cognition and Consciousness Research, University of Salzburg, Salzburg, Austria
    2. Centre for Cognitive Neuroscience Salzburg (CCNS), University of Salzburg, Salzburg, Austria
    Contribution
    Conceptualization, Data curation, Writing – review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0001-5177-4389
  5. Markus Werkle-Bergner

    Center for Lifespan Psychology, Max Planck Institute for Human Development, Berlin, Germany
    Contribution
    Conceptualization, Data curation, Funding acquisition, Methodology, Project administration, Writing – review and editing
    For correspondence
    werkle@mpib-berlin.mpg.de
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-6399-9996

Funding

Deutsche Forschungsgemeinschaft (WE 4269/5-1)

  • Markus Werkle-Bergner

Jacobs Foundation (Early Career Research Fellowship 2017-2019)

  • Markus Werkle-Bergner

Austrian Science Fund (T397-B02)

  • Kerstin Hoedlmoser

Jacobs Foundation (JS1112H)

  • Kerstin Hoedlmoser

Austrian Science Fund (W1233-G17)

  • Michael A Hahn

Austrian Science Fund (P25000)

  • Kerstin Hoedlmoser

German Academic Exchange Service (PRIME Fellowship)

  • Michael A Hahn

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication. Open access funding provided by Max Planck Society.

Acknowledgements

This research was conducted within the project “Lifespan Rhythms of Memory and Cognition” (RHYME, PI: MW-B) at the Max Planck Institute for Human Development (MPIB), Berlin, Germany, and at the Department of Psychology, Laboratory for Sleep, Cognition and Consciousness Research, University of Salzburg, Salzburg, Austria (PI: KH). A-KJ is a fellow of the International Max Planck Research School on the Life Course (LIFE; http://www.imprs-life.mpg.de/en). MW-B received support from the German Research Foundation (DFG, WE 4269/5–1) and the Jacobs Foundation (Early Career Research Fellowship 2017–2019). KH was supported by Austrian Science Fund (T397-B02, P25000), the Jacobs Foundation (JS1112H), and the Centre for Cognitive Neuroscience Salzburg (CCNS). MAH was additionally supported by the Doctoral College “Imaging the Mind” (FWF, Austrian Science Fund W1233-G17) and the PRIME programme of the German Academic Exchange Service (DAAD) with funds from the German Federal Ministry of Education and Research (BMBF). We thank S Wehmeier for her invaluable help with data collection at the MPIB. We are grateful to all members of the RHYME and LIME projects for valuable feedback. We further acknowledge support by the Max Planck Dahlem Campus of Cognition (MPDCC). Finally, we thank all our participants and their families for their time as well as the principals of the schools and the local education authority (Mag. Dipl. Paed. B Heinrich, Prof. Mag. J Thurner) in Salzburg who supported this research.

Ethics

All studies were designed in accordance with the Declaration of Helsinki and approved by the local ethics committee of either the MPIB, Germany (2018_5_Sleep_Conmem), or the University of Salzburg, Austria (EK-435 GZ:16/2014). For the 5- to 6-year-olds, legal guardians gave their written and minors gave their oral informed consent prior to study participation. For the longitudinal study, participants and their legal custodian provided written informed consent before entering the study at every test time point. Similarly, adults provided written informed consent.

Senior and Reviewing Editor

  1. Laura L Colgin, University of Texas at Austin, United States

Reviewer

  1. Hong-Viet Ngo, University of Essex, United Kingdom

Version history

  1. Preprint posted: September 6, 2022 (view preprint)
  2. Received: September 19, 2022
  3. Accepted: October 18, 2023
  4. Version of Record published: November 24, 2023 (version 1)

Copyright

© 2023, Joechner et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.

Metrics

  • 432
    Page views
  • 60
    Downloads
  • 0
    Citations

Article citation count generated by polling the highest count across the following sources: Crossref, PubMed Central, Scopus.

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Open citations (links to open the citations from this article in various online reference manager services)

Cite this article (links to download the citations from this article in formats compatible with various reference manager tools)

  1. Ann-Kathrin Joechner
  2. Michael A Hahn
  3. Georg Gruber
  4. Kerstin Hoedlmoser
  5. Markus Werkle-Bergner
(2023)
Sleep spindle maturity promotes slow oscillation-spindle coupling across child and adolescent development
eLife 12:e83565.
https://doi.org/10.7554/eLife.83565

Share this article

https://doi.org/10.7554/eLife.83565

Further reading

    1. Neuroscience
    Eyal Y Kimchi, Anthony Burgos-Robles ... Kay M Tye
    Research Article

    Basal forebrain cholinergic neurons modulate how organisms process and respond to environmental stimuli through impacts on arousal, attention, and memory. It is unknown, however, whether basal forebrain cholinergic neurons are directly involved in conditioned behavior, independent of secondary roles in the processing of external stimuli. Using fluorescent imaging, we found that cholinergic neurons are active during behavioral responding for a reward – even prior to reward delivery and in the absence of discrete stimuli. Photostimulation of basal forebrain cholinergic neurons, or their terminals in the basolateral amygdala (BLA), selectively promoted conditioned responding (licking), but not unconditioned behavior nor innate motor outputs. In vivo electrophysiological recordings during cholinergic photostimulation revealed reward-contingency-dependent suppression of BLA neural activity, but not prefrontal cortex. Finally, ex vivo experiments demonstrated that photostimulation of cholinergic terminals suppressed BLA projection neuron activity via monosynaptic muscarinic receptor signaling, while also facilitating firing in BLA GABAergic interneurons. Taken together, we show that the neural and behavioral effects of basal forebrain cholinergic activation are modulated by reward contingency in a target-specific manner.

    1. Neuroscience
    Olgerta Asko, Alejandro Omar Blenkmann ... Anne-Kristin Solbakk
    Research Article Updated

    Orbitofrontal cortex (OFC) is classically linked to inhibitory control, emotion regulation, and reward processing. Recent perspectives propose that the OFC also generates predictions about perceptual events, actions, and their outcomes. We tested the role of the OFC in detecting violations of prediction at two levels of abstraction (i.e., hierarchical predictive processing) by studying the event-related potentials (ERPs) of patients with focal OFC lesions (n = 12) and healthy controls (n = 14) while they detected deviant sequences of tones in a local–global paradigm. The structural regularities of the tones were controlled at two hierarchical levels by rules defined at a local (i.e., between tones within sequences) and at a global (i.e., between sequences) level. In OFC patients, ERPs elicited by standard tones were unaffected at both local and global levels compared to controls. However, patients showed an attenuated mismatch negativity (MMN) and P3a to local prediction violation, as well as a diminished MMN followed by a delayed P3a to the combined local and global level prediction violation. The subsequent P3b component to conditions involving violations of prediction at the level of global rules was preserved in the OFC group. Comparable effects were absent in patients with lesions restricted to the lateral PFC, which lends a degree of anatomical specificity to the altered predictive processing resulting from OFC lesion. Overall, the altered magnitudes and time courses of MMN/P3a responses after lesions to the OFC indicate that the neural correlates of detection of auditory regularity violation are impacted at two hierarchical levels of rule abstraction.