The infraslow fluctuation of sigma power during sleep and its links to markers of arousal and memory reactivation across development Authors: Maria E Dimitriades, Alejandro Osorio-Forero, Sara Fattinger, Simone von Arx, Salome Kurth, Fiona Pugin, Valeria Jaramillo, Carina Volk, Elena Krugliakova, Melanie Furrer, Sven Leach, Peter Achermann, Miriam Gerstenberg, Reto Huber Published: 2026 ABSTRACT AbstractSleep is both a state of disconnection from the environment and a critical period for restoration. But how does sleep balance responsiveness with the protection of key functions? The infraslow fluctuation of sigma power (ISFS)—the clustering of sleep spindles over 10–100 s—is thought to regulate this trade-off in rodents by alternating between fragile periods of high arousability and protected periods of spindle-rich stability. However, the organization of arousal and memory reactivation markers within the human ISFS and its conservation in younger ages remain unclear. This study characterizes the ISFS from childhood to young adulthood (N = 154; ages 8–26), examining its relationship with electrophysiological markers of arousal (i.e., microarousals) and memory reactivation (i.e., slow wave-sleep spindle coupling). Results indicate that the ISFS is present across all ages, with frequency, variability, and strength increasing from early to late adolescence. Notably, electrophysiological markers of arousal and memory reactivation are organized within the spindle-rich ISFS peak. The consistent presence and organization of the ISFS suggest it may be intrinsic to sleep, with adolescence marking a dynamic window. These insights may guide interventions to promote healthier sleep across development.Supplementary InformationThe online version contains supplementary material available at https://doi.org/10.1038/s41598-026-58423-z.Keywords: Sleep, Infraslow, Sleep Spindles, Development, Arousal, Memory, high-density EEGSubject terms: Locus coeruleus, Non-REM sleep Introduction Sleep comes with considerable evolutionary risks—reduced responsiveness and increased vulnerability to potential dangers. A growing hypothesis, based on rodent studies, suggests that sleep is not merely a vulnerable state but rather orchestrates a balance between environmental awareness and critical protected periods for learning and memory1,2. Recent research highlighted the importance of activity occurring on an infraslow time scale, over the course of 10 to 100 s, in achieving this balance2,3. In particular, sleep spindles, brief oscillatory bursts in the sigma range (10–16 Hz) that occur during non-rapid eye movement (NREM) sleep, do not occur uniformly throughout sleep, but instead fluctuate on an infraslow time scale4. In rodents, the infraslow fluctuation of sigma power (ISFS) is driven by the locus coeruleus, a brainstem nucleus that regulates arousal through norepinephrine release1. The rodent ISFS segments NREM sleep into alternating fragile and protected periods1,2,5. Fragile periods demonstrate decreasing sleep spindle activity and increasing locus coeruleus activity, which is associated with a reduced arousal threshold in response to external stimuli. In contrast, protected periods show increasing sleep spindle activity, an elevated arousal threshold, and markers of memory reactivation (i.e., ripple activity). Researchers have interpreted this dynamic as an evolutionary trade-off, where fragile periods prioritize sensory vigilance and environmental awareness, while protected periods favor internal processing, including memory reactivation, by minimizing external disturbances2. Yet, the organization of markers of arousal and memory reactivation remain understudied in humans6. While studies in adults have confirmed the presence of the ISFS in humans2,6–8, its trajectory and functional significance remain largely unexplored in child and adolescent populations. Childhood and adolescence represent critical periods of intense learning and brain maturation, during which sleep structure and electrophysiological characteristics, including sleep spindles, undergo pronounced developmental changes relative to both early childhood and adulthood. Across this period, sleep becomes progressively more adult-like, with decreases in total sleep time and the proportion of N3 sleep, reductions in electroencephalographic (EEG) amplitude, and increases in spindle frequency and density, alongside a higher incidence of sleep problems, such as impaired sleep continuity9–16. Given the relationship between sleep, arousal, and memory reactivation during these developmental stages, examining the ISFS may provide insights into how these processes interact to support learning and cognitive development. This research aims to analyze sleep EEG data from children, early adolescents, late adolescents, and young adults to: (1) assess the presence and parameters of the ISFS from childhood to adulthood and (2) investigate how the ISFS relates to electrophysiological markers of arousal and memory reactivation throughout this developmental trajectory. Materials and methods Participants EEG sleep data were pooled from studies collected between 2008 and 202117–22 conducted in the sleep laboratory of the University Children’s Hospital Zurich. Individuals were aged 8 to 26 years (N = 154, 16.3 ± 5.1 years, 38.3% female and 61.7% male) and four age groups were derived: children (8 to 11 years old; N = 39, 10.4 ± 1.1 years, 41.0% female and 59.0% male), early adolescents (12 to 15 years old; N = 45, 14.0 ± 1.2 years, 13.3% female and 86.7% male), late adolescents (16 to 20 years old; N = 32, 18.4 ± 1.7 years, 53.1% female and 46.9% male), and young adults (21 to 26 years old; N = 38, 23.5 ± 1.6 years, 55.3% female and 44.7% male). Individuals were excluded if they had a diagnosed sleep disorder, a history of mental health disorders, consumed psychotropic medications, and/or did not abide by a regular sleep schedule the week prior to the recording. Written informed consent and/or assent was obtained from all participants and/or legal guardians, respectively. Data collection was approved by the cantonal ethics committee, and all study procedures were performed according to the Declaration of Helsinki. Electroencephalographic and electromyographic data EEG sleep data were recorded using a 128-channel system (Electrical Geodesic Sensory Net; sampling rate: 500–1000 Hz). Sleep staging and artifact rejection were available from previous analyses17–22. Sleep staging was performed on 20 s epochs based on the American Association of Sleep Medicine (AASM) criteria by two sleep experts23. The earliest datasets were scored using 20 s epochs in line with AASM criteria at the time, and subsequent datasets followed this epoch length to maintain methodological consistency across the cohort. Semi-automatic rejection of artifacts was performed based on power in two frequency bands, delta (0.5–4.5 Hz) and beta (20–30 Hz), as described by Huber and colleagues24 and Leach and colleagues25. Channels with artifacts in 90% or more of their NREM epochs were considered of poor quality. On average, this resulted in 2.09 channels being rejected per participant (range: 0–11 channels). Poor quality channels were interpolated from neighboring electrodes. Artifact rejection targeted both non-physiological artifacts (hardware- or environmental-related disturbances such as high-amplitude deflections or cable movement) and physiological artifacts (arising from eye and body movements, often including microarousals)25. For the multichannel analyses, only epochs that were free of both non-physiological and physiological artifacts across 109 electrodes (excluding the electrodes on the edge of the net, which cover neck and face muscles) were included25. For the single-channel analyses, epochs initially marked as artifacts in the given channel were visually re-evaluated, and only non-physiological artifacts and eye movements were excluded. For the multichannel analyses, EEG data for the infraslow analyses were low-pass filtered (-3 dB cut-off: 39.25 Hz; Hamming windowed sinc Finite Impulse Response (FIR) filter from eeglab, filter order: 184), downsampled (200 Hz), and high-pass filtered (-3 dB cut-off = 0.27 Hz, Kaiser Window FIR filter, filter order: 2390)26. The data were re-referenced to the average of 109 electrodes (same as above)25,27. For the single channel analyses, EEG data for the infraslow, sleep spindle, and microarousal analyses were low-pass filtered (-3 dB cut-off: 39.25 Hz; Hamming windowed sinc FIR filter from eeglab, filter order: 184), downsampled (200 Hz), and high-pass filtered (-3 dB cut-off = 0.27 Hz, Kaiser Window FIR filter, filter order: 2390)26. The infraslow and sleep spindle data were re-referenced to the average of 109 electrodes (same as above), to be consistent with the multichannel analyses25,27. The microarousal data were re-referenced to the contralateral mastoid, according to AASM criteria23. Additionally, electromyography (EMG) data were used for the detection of microarousals. EMG signals were low-pass filtered (-6dB cut-off: 101.25 Hz; Hamming windowed sinc FIR filter from eeglab, filter order: 75), downsampled (200 Hz), and high-pass filtered (-6 dB cut-off: 9.63 Hz; Kaiser Window FIR filter, filter order: 301) according to AASM criteria23,26. EEG data for slow wave detection were instead band-pass filtered (-3 dB cut-offs: 0.34 and 7.1 Hz; Chebyshev Type II IIR filter; filter order: 2 using filtfilt), downsampled (200 Hz), and re-referenced to the mean of the mastoids28. Analyses of the infraslow fluctuation of sigma power Bouts of NREM Stage 2 (N2) sleep data that lasted for at least 280 s and were uninterrupted by artifacts or transitions to other vigilance states were isolated (demonstrated in Fig. 1A). The minimum bout length of 280 s was chosen to ensure two full cycles of the lowest frequency of interest (0.0075 Hz). Our analyses focused on N2 sleep, where the ISFS is reported to be most prominent2. All subsequent analyses were performed on the aforementioned bouts. Overview of Analyses Assessing the Infraslow Fluctuation of Sigma Power (A) Top: Example hypnogram with bouts of consecutive, artifact-free non-rapid eye movement (NREM) stage 2 sleep lasting at least 280 s highlighted in green. Bout data will be used in all subsequent analyses. Bottom: Time series data is shown in grey with detected sleep spindles highlighted in blue. Sleep spindles are identified using an automatic detection method. (B) A wavelet transform is performed on the time series data (corresponds to the data seen in A; sleep spindles marked as blue dots) to derive the sigma power time course (the mean of the absolute value of the sigma power time courses derived from 10 to 16 Hz) which is shown in grey. (C) Outline of the two key analyses of the infraslow fluctuation of sigma power (ISFS) to be performed on the sigma power time course as presented in B: (i) Left: The relative spectral power (the mean spectral power derived over all bouts for each channel and subject) is shown in grey. The relative spectral power between 0.06 to 0.1 Hz, indicated by a red horizontal line, is subtracted for baseline correction. Right: The resulting relative spectral power following baseline correction is shown as a grey line. A Gaussian function (Gaussian 1) is fit to the relative spectral power, shown in dark purple. If the peak power value surpasses the threshold, shown with a dotted line and calculated as 1.5 times the standard deviation of the gaussian, the presence of the ISFS is detected and parameters are further estimated: peak frequency (peak of Gaussian; marked with a black dot), bandwidth (± 1 standard deviation from the peak frequency; black line), and area under the curve (total power within ± 1 standard deviation from the peak frequency; shaded in light purple). (ii) Top: The grey line represents the sigma power time course of part of one bout, with detected sleep spindles marked as blue dots (corresponding to B but over a longer time period to display the ISFS). The black line represents the filtered sigma power time course. Bottom: The filtered sigma power time course is standardized by subtracting the mean and dividing by the standard deviation before detecting peaks, troughs, and zero-crossings. The ISFS is further divided into eight phase bins; 1–4 denote the negative half wave and 5–8 denote the positive half wave. Any signal that could not be fully estimated, i.e., that did not occur on infraslow time scale (25 to 100 s) and/or was at the start or end of the bout and therefore did not contain both a peak and a trough, was labeled “Not ISFS”. The sigma power time course was extracted from the N2 sleep bout data using a Mother Gabor-Morlet wavelet, following the method previously applied in mice1(cycles: 4, frequency resolution: 0.2 Hz; demonstrated in Fig. 1B). Considering that the frequency of sleep spindles increases across development and is generally higher in central compared to frontal regions4,15, we selected a broad frequency range, 10 to 16 Hz, to capture these differences in subsequent analyses. The absolute value of each sigma power time course derived from the bout, corresponding to the 31 input frequencies (ranging from 10 to 16 Hz in 0.2 Hz steps), was first calculated. The mean sigma power time course was then computed and used for subsequent analyses. To recenter the data for each bout around zero, the mean of that bout was subtracted from the sigma power time course. Spectral analysis of the sigma power time course The FFT was applied to the sigma power time course of each bout (demonstrated in Fig. 1C). To account for the varying lengths of the bouts (defined as lasting 280 s or more), the spectral power was adjusted by dividing the FFT output by the length of the corresponding bout. This adjustment ensures that the spectral power accurately corrects for the bout duration and allows for meaningful comparisons across different bouts. Some bouts exhibited a low-frequency component in the spectral power ranging from 0 to 0.004 Hz, which was not systematically linked to specific participants, channels, or sleep stages. This low-frequency component was occasionally removed with linear detrending and is believed to stem from the extraction of the sigma power time course. When the spectral power values in this low-frequency range exceeded the peak power detected outside this range (0.004–0.1 Hz), the power values from 0 to 0.004 Hz were excluded. This exclusion was applied before calculating the mean spectral power across all bouts, done separately for each channel and participant. Relative power was calculated as the spectral power of each bout divided by the mean power within that same bout for each channel and participant; only power values from 0 to 0.1 Hz were considered. Absolute power was not reported in this project due to the higher EEG amplitude values typically observed in children and adolescents17, which significantly impact this measure. To detect a peak in the infraslow range and parameterize the curve, a Gaussian function (Gaussian 1) was fitted to the spectral power. Prior to fitting, the mean power value from 0.06 to 0.1 Hz was subtracted from the signal as a horizontal baseline. This correction was applied to remove any offset, as this frequency range visually tends to plateau before the spectrum rises to a peak near 0.25 Hz, as previously described by Achermann and Borbély29. Further, the frequency range of interest for detecting the peak of the ISFS was set between 0.01 and 0.04 Hz as this range typically contains the infraslow spectral peak1,2. When the peak power value surpassed the threshold, defined as 1.5 times the standard deviation of the Gaussian curve, the signal in the given channel was considered to contain a detectable ISFS. The ISFS was characterized by three parameters: peak frequency (representing the most common period of the ISFS), bandwidth (representing the variability in period of the ISFS, calculated as ± 1 standard deviation away from the peak frequency), and area under the curve (representing the strength of the ISFS, calculated as the total power within ± 1 standard deviation away from the peak frequency). The bandwidth and area under the curve are defined by the same range of frequencies making these measures interrelated. To analyze age-related differences in the spectral parameters, the mean value across all electrodes was computed, excluding electrodes without a detectable ISFS. Additionally, to assess the topographical distribution of the ISFS spectral parameters independently of age-related changes, relative values were calculated by dividing the value of each electrode by the mean value of all electrodes for that participant30. The spectral analysis of the ISFS was performed on the total and reduced number of bouts (see Table 1) to ensure comparability between the age groups. Table 1Parameters of sleep architecture.ParametersCHEALAYAMean (SD)Mean (SD)Mean (SD)Mean (SD)Corrected p-values and Direction All Night Sleep N1 (min)27.39 (13.14)22.33 (13.42)27.03 (18.39)25.76 (12.23)0.24N2 (min)210.81 (45.43)213.38 (40.43)214.59 (32.35)224.32 (29.59)0.45N3 (min)139.62 (49.86)138.76 (42.87)85.93 (28.45)79.45 (32.04)LA < CH (< 0.0001) YA < CH (< 0.0001) LA < EA (< 0.0001) YA < EA (< 0.0001)REM (min)92.23 (23.31)83.74 (27.22)88.61 (30.28)98.14 (23.74)0.12Total Sleep (min)470.05 (43.37)458.21 (52.66)416.12 (53.36)427.68 (30.96)LA < CH (< 0.0001) YA < CH (< 0.001) LA < EA (< 0.01) YA < EA (0.01) Total Bout Sleep Bout Number11.36 (4.78)11.47 (4.47)11.97 (4.32)8.97 (4.43)0.05Mean Bout Length (s)495.23 (95.24)470.68 (102.03)489.28 (65.72)458.39 (113.16)0.12Total Duration of Bouts (min)97.59 (48.55)91.36 (43.21)99.23 (41.93)72.56 (43.33)0.07Mean Relative Location (%)56.43 (7.30)56.02 (8.27)52.87 (6.58)49.36 (11.68)YA < CH (0.02) YA < EA (0.03)Proportion of all N244.10 (17.79)41.97 (16.88)45.69 (16.06)32.18 (18.87)YA < CH (0.03) YA < LA (0.01) Reduced Bout Sleep Bout Number9.51 (4.49)9.51 (4.38)9.97 (4.32)8.97 (4.43)0.96Mean Bout Length (s)489.27 (89.10)467.63 (106.58)485.82 (70.97)458.39 (113.16)0.18Total Duration of Bouts (min)80.79 (42.83)76.47 (41.62)82.22 (40.12)72.56 (43.33)0.78Mean Relative Location (%)50.15 (7.85)49.44 (9.10)45.61 (7.76)49.36 (11.68)0.16Proportion of all N236.24 (15.78)34.70 (16.60)37.63 (15.62)32.18 (18.87)0.55CH: Children; EA: Early adolescents; LA: Late adolescents; YA: Young adults; SD: Standard Deviation; NREM: Non-Rapid Eye Movement (N1, N2, N3); REM: Rapid Eye Movement; min: minutes; s: seconds; %: percent. Bout: N2 sleep data that lasts for at least 280 s and is uninterrupted by artifacts or transitions to other vigilance states that was used for the spectral analyses of sigma power time course. Total Bout Sleep: The maximum number of bouts are taken from every individual. Reduced Bout Sleep: The last two bouts are excluded from the child, early adolescent, and late adolescent groups to make the bout parameters comparable. Mean Relative Location: the mean across all bout midpoints divided by the total length of the recording. Proportion of all N2: the total duration of bout data is divided by the total amount of N2 sleep. Corrected p-values and the direction of the difference between age groups is plotted with significantly different p-values written in bold. Parameters of sleep architecture. CH: Children; EA: Early adolescents; LA: Late adolescents; YA: Young adults; SD: Standard Deviation; NREM: Non-Rapid Eye Movement (N1, N2, N3); REM: Rapid Eye Movement; min: minutes; s: seconds; %: percent. Bout: N2 sleep data that lasts for at least 280 s and is uninterrupted by artifacts or transitions to other vigilance states that was used for the spectral analyses of sigma power time course. Total Bout Sleep: The maximum number of bouts are taken from every individual. Reduced Bout Sleep: The last two bouts are excluded from the child, early adolescent, and late adolescent groups to make the bout parameters comparable. Mean Relative Location: the mean across all bout midpoints divided by the total length of the recording. Proportion of all N2: the total duration of bout data is divided by the total amount of N2 sleep. Corrected p-values and the direction of the difference between age groups is plotted with significantly different p-values written in bold. Temporal dynamics analysis of the sigma power time course The sigma power time course of each bout was also analyzed to study the ISFS over time in one central channel (C4; demonstrated in Fig. 1C). The sigma power time course was low-pass filtered (-3 dB cut-off = 0.04 Hz, Butterworth Infinite Impulse Response (IIR) filter, filter order: 5 using filtfilt, Tukey window with a cosine fraction of 0.1) and then standardized by subtracting the mean and dividing by the standard deviation. All peaks, troughs, and zero-crossings (when the filtered sigma power time course crossed the zero-line corresponding to the mean of the signal) were identified. The presence of the ISFS was identified when two negative zero-crossings were separated by 25 to 100 s (i.e., 0.01–0.04 Hz, corresponding to our frequency range of interest mentioned above). To assess where electrophysiological characteristics (defined below) occurred within the ISFS, the filtered sigma power time course defining the ISFS was divided into eight phase bins. The start of the ISFS (first negative zero-crossing) until the midpoint (positive zero-crossing) is the negative half wave of the ISFS. It is further divided by the trough: the mean sample between the start and the trough separates bins 1 and 2 and the mean sample between the trough and the midpoint separates bins 3 and 4. Moreover, the midpoint of the ISFS until the end of the ISFS (second negative zero-crossing) is the positive half wave of the ISFS. It is further divided by the peak: the mean sample between the midpoint and the peak separates bins 5 and 6 and the mean sample between the peak and the end separates bins 7 and 8. Any signal that could not be fully estimated, i.e., that did not occur on infraslow time scale (25 to 100 s) and/or was at the start or end of the bout and therefore did not contain both a peak and a trough, was labeled “Not ISFS”. Detection of electrophysiological characteristics We detected spontaneous microarousals with an established algorithm31,32. In brief, this algorithm uses EEG and EMG data to detect microarousals by identifying increases in baseline alpha and/or beta activity above a relative threshold, signaling the presence of an arousal event. Detected microarousals with a duration between three and 10 s were considered valid, in accordance with AASM criteria. The algorithm includes an automatic spindle-exclusion step, in which candidate microarousals are re-evaluated using a sliding window to identify and mask sigma-band (12–15 Hz) activity; only events containing a contiguous non-spindle segment of at least three seconds are retained. As our sample included younger individuals, who typically have lower sleep spindle frequencies15, we repeated the analysis examining the distribution of microarousals within the ISFS using only beta-band microarousals to avoid any overlap with spindles. Additionally, we repeated the analysis with a reduced minimum duration (1.5 s) of microarousals to capture a broader range of responses that have been previously reported in younger individuals33–35. Examples of microarousals detected in the presence or absence of spindles are shown in Supplementary Fig. 1. The onset of the given microarousal was taken as the sample of the event. Sleep spindles were identified using an established amplitude-threshold algorithm36. In short, N2 sleep bout data were isolated and band-pass filtered (-3 dB cut-offs: 9.96 and 17.5 Hz; Chebyshev Type II IIR filter; filter order: 6 using filtfilt). Given the variability in signal amplitude across electrodes, relative upper and lower thresholds were established for each channel and participant. Sleep spindles were detected in a given channel when the amplitude exceeded an upper threshold of five times the average signal amplitude and until the instances where the signal surrounding the maximum peak fell below two times the average signal amplitude for the given channel. The minimal duration of a sleep spindle was set to be 0.3 seconds37. The maximum amplitude of the sleep spindle was taken as the sample of the event. Slow waves were detected using an established algorithm28. Bouts of N2 sleep data were band-pass filtered (see Electroencephalographic and Electromyographic Data Section for details) and all peaks, troughs, and zero-crossings were identified. Slow waves were identified as the time between two negative zero-crossings, separated by intervals ranging from 0.5 to two seconds. To exclude outliers, the bottom 10% and top 1% of peak-to-trough amplitudes were removed. Further, in order to target high-amplitude slow waves, only those within the top 80th percentile of peak-to-trough amplitudes were considered for analysis as previously reported28. Slow wave-sleep spindle coupling was further defined when the maximum amplitude of a detected sleep spindle fell within the up-phase of a detected high-amplitude slow wave. The first negative zero-crossing was taken as the sample of the event. To assess the distribution of microarousal onsets within each phase bin of the ISFS, the number of detected microarousals in each phase bin of the ISFS was calculated as a percentage of the total number of microarousals detected for each participant; only individuals with at least five microarousals were included. The same calculation was done to assess the distribution of sleep spindles, high-amplitude slow waves, and high-amplitude slow wave-sleep spindle coupling within the phase bins of the ISFS. The relative coupling of high-amplitude slow waves and sleep spindles was also quantified by dividing the number of sleep spindles coupled to a high-amplitude slow wave within a given phase bin of the ISFS by the total number of sleep spindles in that phase bin of the ISFS. The temporal distribution of high-amplitude slow waves and sleep spindles around microarousals was analyzed in 0.5-second bins from two seconds before to eight seconds after the onset of the microarousal. This time window was chosen because the average duration of a microarousal is approximately eight seconds (see Supplementary Table 1). These data were compared to equivalent 0.5-second control bins drawn from NREM periods that did not fall within the microarousal window, providing a larger pool of control data. Event occurrence is expressed as the number of events in each 0.5 s time bin divided by the total number of events for that participant in the − 2 to + 8 s window, multiplied by 100, and then averaged across participants within each age group. Additionally, the coupling of sleep spindles occurring within the 0.5 to two seconds after the onset of the microarousal was compared to any sleep spindle not occurring during this window, providing a larger pool of data. Statistical analysis All statistical analyses were conducted in MATLAB R2022a. Normality was assessed using quantile-quantile plots and Shapiro-Wilk tests. Kruskal-Wallis tests, followed by Tukey-Kramer post-hoc tests, were used to compare sleep architecture parameters and the parameters of electrophysiological characteristics occurring during N2 bout data for the four age groups. Additionally, they were used to compare the distribution of electrophysiological characteristics within the phase bins of the ISFS. Mann-Whitney U tests with false discovery rate (Benjamini-Hochberg) correction were conducted to compare the percentage of sleep spindles coupling within high-amplitude slow waves across conditions. High-amplitude slow wave (defined as the first negative zero-crossing) and sleep spindle (defined as the maximum amplitude) occurrence was quantified in non-overlapping 0.5 s bins from − 2 to + 8 s around microarousal onset. Wilcoxon signed-rank tests, corrected for multiple comparisons using the Benjamini-Hochberg false discovery rate procedure, compared these bins to equivalent 0.5 s control bins that did not fall within the microarousal window. One-way ANOVAs were used to compare the ISFS spectral parameters across channels, with Tukey-Kramer post-hoc tests for multiple comparisons. To account for potential sex effects given unequal group ratios, the ANOVAs were repeated for the ISFS spectral parameters including sex as a covariate (Type III sum of squares), and Kruskal-Wallis tests for the distribution of electrophysiological characteristics within the phase bins of the ISFS were repeated separately for females and males. Electrode-wise topographical differences were assessed using one-way ANOVAs, followed by cluster-based permutation testing. A p-value less than 0.05 was considered significant. Results Sleep architecture parameters Sleep architecture parameters have been previously published17–22. All-night sleep parameters showed that the total amount of NREM sleep stage N1 and N2 and Rapid Eye Movement (REM) sleep were comparable between groups (Table 1). As expected, significant increases in N3 sleep and total sleep time were detected in the child and early adolescent groups when compared to the late adolescent and young adult groups. To analyze data related to the ISFS, continuous N2 sleep bout data were examined. The number of bouts, mean bout length, and the total duration of N2 sleep bout data were similar across age groups. The presence and spectral parameters of the infraslow fluctuation of sigma power Spectral analysis of the sigma power time course revealed that the ISFS was detectable in children, early adolescents, late adolescents, and young adults (Fig. 2A). All individuals had a detectable ISFS in at least 20% of electrodes (mean ± standard deviation in % of electrodes with a detectable ISFS: children: 74.74 ± 21.08, early adolescents: 75.66 ± 21.83, late adolescents: 82.05 ± 18.14, young adults: 80.87 ± 24.85). The mean value of each ISFS spectral parameter, peak frequency, bandwidth, and area under the curve, was calculated across channels. All ISFS spectral parameters showed significant differences between age groups (Fig. 2B): peak frequency was significantly higher in the young adult group compared to the child and early adolescent groups (Tukey-Kramer post-hoc: children vs. young adults, p = 0.003; early adolescents vs. young adults, p = 0.020; percent change: children vs. young adults = 22.37, early adolescents vs. young adults = 17.24; F(3,150) = 4.60, p = 0.004, ηp2 = 0.08), bandwidth was significantly higher in the late adolescent group compared to the child group (Tukey-Kramer post-hoc: children vs. late adolescents, p < 0.001; percent change: children vs. late adolescents = 52.59; F(3,150) = 6.32, p < 0.001, ηp2 = 0.11), and area under the curve was significantly higher in the late adolescent and young adult groups compared to the child group (Tukey-Kramer post-hoc: children vs. late adolescents, p = 0.002; children vs. young adults, p = 0.007; percent change: children vs. late adolescents = 45.22, children vs. young adults = 38.47; F(3,150) = 5.38, p = 0.002, ηp2 = 0.10). Additionally, when sex was included as a covariate, the main effects of age group on all ISFS spectral parameters remained significant (PF: F(3,149) = 3.52, p = 0.017, ηp2 = 0.07; BW: F(3,149) = 7.16, p < 0.001, ηp2 = 0.13; AUC: F(3,149) = 5.47, p = 0.001, ηp2 = 0.10). A sex effect was observed for bandwidth (F(1,149) = 6.85, p = 0.009, ηp2 = 0.04; males had higher bandwidth than females), whereas sex effects were nonsignificant for peak frequency and area under the curve (PF: F(1,149) = 2.55, p = 0.112, ηp2 = 0.02; AUC: F(1,149) = 0.36, p = 0.550, ηp2 = 0.002). These results indicate that age-related differences in ISFS properties remain robust after accounting for sex, although overall ISFS bandwidth showed a modest sex-related effect, characterized by higher bandwidth in males than in females. The presence and spectral parameters of the infraslow fluctuation of sigma power across development. (A) Bold lines represent the smoothed mean spectral power of the infraslow fluctuation of sigma power (ISFS) averaged across all electrodes for each age group (CH: children; EA: early adolescents; LA: late adolescents; YA: young adults). The standard deviation for each age group is shown in paler colors with dotted lines. (B) The mean spectral parameters of the ISFS (peak frequency, bandwidth, area under the curve) averaged across all electrodes are shown for individuals (denoted by dots) for all age groups. Significant differences between age groups are indicated by an asterisk(s). (C) Rows represent the relative values of each spectral parameter of the ISFS, while columns correspond to different age groups. Relative values were calculated by dividing the value of each electrode by the mean value of all electrodes for that participant to assess the topographical distribution of the ISFS spectral parameters independently of age-related changes. A circle in the top row highlights a cluster of frontal electrodes that show significant differences across the age groups. Topographical analysis of the ISFS spectral parameters found that most electrodes showed no significant differences across age groups (Fig. 2C). However, regarding peak frequency, a frontal cluster of seven electrodes revealed significantly higher values in the young adult group compared to the child and early adolescent groups (cluster mean ± standard deviation: children = 0.94 ± 0.02, early adolescents 0.98 ± 0.01, young adults 1.11 ± 0.04; young adult vs. child p = 0.009; young adult vs. early adolescent p = 0.027; ηp2 ranges from 0.08 to 0.13 in the cluster of significant electrodes), indicating higher peak frequencies frontally with age. Peak frequency and area under the curve showed local minima and maxima in central regions, respectively, while bandwidth displayed no clear topographical pattern. All the above analyses could be confounded by significant age-related differences in the relative location of each bout (within the total recording) and the proportion of N2 bout data (relative to total N2 sleep; Table 1). Thus, in order to compare a similar distribution and proportion of bout sleep data, the two final bouts were removed from the child, early adolescent, and late adolescent groups, which resulted in no statistical difference between groups for the bout parameters (Table 1). The repeated spectral analyses revealed qualitatively similar age- and topographical-related changes of the ISFS spectral parameters (data not shown). Organization of electrophysiological characteristics within the infraslow fluctuation of sigma power All subsequent analyses focus on microarousals, sleep spindles, high-amplitude slow waves, and high-amplitude slow wave-sleep spindle coupling in a central electrode, chosen for its robust ISFS across all age groups as indicated by a high area under the curve (reported in Fig. 2C; details on the bout data used in these analyses can be found in Supplementary Table 1). To assess the organization of electrophysiological characteristics within the ISFS, the ISFS was divided into eight phase bins. The amount of data in the positive and negative half waves of the ISFS was comparable (49–51%), as well as the amount of data in each of the eight phase bins of the ISFS and the data where the ISFS was not able to be estimated (Phase bins 1–8: 10–11%; Not ISFS: 12–15%) across age groups. This indicates that the distribution of electrophysiological characteristics across each half wave, phase bin, or Not ISFS was not biased by the duration of the time frame analyzed. When investigating the distribution of the onset of microarousals within the ISFS, only participants with at least five detected microarousals were included (children = 29; early adolescents = 32; late adolescents = 27; young adults = 31). The threshold corresponded to the 25th percentile of the total number of detected microarousals across participants (see Supplementary Fig. 2 for results using alternative thresholds). The microarousal number, density, and duration in the N2 bout data was similar across age groups (see Supplementary Table 1). The distribution of microarousals significantly differed between phase bins of the ISFS for all age groups (Fig. 3A; children: H(8) = [35.19], p < 0.0001; early adolescents: H(8) = [64.00], p < 0.0001; late adolescents: H(8) = [100.56], p < 0.0001; young adults: H(8) = [122.14], p < 0.0001). The highest percentage of microarousals occurred around the peak of the ISFS (corresponding to phase bin 5, 6, and/or 7) compared to phase bin(s) occurring in the negative half wave of the ISFS for all age groups. This analysis was repeated using only beta band criteria for microarousals and reduced minimum duration criteria for microarousals (1.5 instead of three seconds). Both additional analyses revealed the same distribution of microarousals, with the majority occurring at the peak of the ISFS (see Supplementary Fig. 3). Distribution of Electrophysiological Characteristics within the Infraslow Fluctuation of Sigma Power (A) The distribution of the onset of microarousals across bins of the infraslow fluctuation of sigma power (ISFS) and in periods where the ISFS could not be estimated (denoted as “Not ISFS”) is displayed. Participants are excluded from this analysis if they did not have at least five microarousals. (B) The distribution of sleep spindles across phase bins of the ISFS and Not ISFS is displayed. (C) The distribution of high-amplitude slow waves across phase bins of the ISFS and Not ISFS is displayed. (D) The distribution of high-amplitude slow wave-sleep spindle coupling across phase bins of the ISFS and Not ISFS is displayed. All electrophysiological characteristics are calculated by dividing the number of electrophysiological characteristics in each phase bin of the ISFS by the total number of electrophysiological characteristics for each participant, with the mean across participants displayed. Each column represents an age group, with significant differences between phase bins marked by asterisk(s). For groups and parameters with many significant differences between groups, grey arrows are used to show groups of phase bins that are significantly higher when compared to other phase bins (denoted in grey brackets); the range of p-values written above them in grey. The bold black wave illustrates the relative position of each phase bin within the ISFS. The distribution of sleep spindles, high-amplitude slow waves, and high-amplitude slow wave-sleep spindle coupling also differed between phase bins of the ISFS for most age groups (see Supplementary Table 1 for a quantitative description of these electrophysiological characteristics). Per definition, sleep spindles occurred most in the positive half wave of the ISFS (corresponding to phase bins 5–8; peak: 6–7) when compared to phase bins of the negative half wave of the ISFS (corresponding to phase bins 1–4; trough: 2–3) for all age groups (Fig. 3B; children: H(8) = [308.18]; early adolescents: H(8) = [351.42]; late adolescents: H(8) = [244.39]; young adults: H(8) = [300.19]; all p-values were < 0.0001). Within the ISFS, more high-amplitude slow waves occurred in and around the peak compared to the negative half wave for all age groups (Fig. 3C; children: H(8) = [249.57]; early adolescents: H(8) = [267.77]; late adolescents: H(8) = [267.77]; young adults: H(8 = [247.19]; all p-values were < 0.0001). Subsequently, the coupling of high-amplitude slow waves with sleep spindles also occurred more in the peak of the ISFS when compared to phase bins in the negative half wave of the ISFS for all age groups (Fig. 3D; children: H(8) = [250.27]; early adolescents: H(8) = [281.70]; late adolescents: H(8) = [200.37]; young adults: H(8) = [257.72]; all p-values were < 0.0001). Additionally, the relative coupling, i.e., the percent of sleep spindles that coupled with high-amplitude slow waves in a given phase bin of the ISFS after dividing by the total number of sleep spindles in the given phase bin, revealed that more coupling occurred in the peak of the ISFS when compared to the trough of the ISFS for the late adolescent and young adult groups (late adolescents: H(8) = [22.65], p = 0.004; young adults: H(8) = [32.46], p < 0.0001). In summary, sleep spindles, high-amplitude slow waves, and high-amplitude slow wave-sleep spindle coupling were more likely to occur in the positive half wave of the ISFS when compared to the negative half wave of the ISFS. Further, performing this analysis separately for females and males (data not shown) also resulted in the microarousals, sleep spindles, high-amplitude slow waves, and coupling events occurring more in the positive half wave of the ISFS for both groups. To further quantify the temporal alignment of electrophysiological events within the ISFS, we calculated the proportion of events occurring during the positive half-wave, negative half-wave, and outside the ISFS (Supplementary Table 2). As expected, all electrophysiological characteristics occurred most frequently during the positive half-wave of the ISFS (57–73%), followed by the negative half-wave (18–33%), and least frequently outside the ISFS (8–10%). High-amplitude slow wave and sleep spindle occurrence and high-amplitude slow wave-sleep spindle coupling following the onset of microarousals To determine if high-amplitude slow waves and sleep spindles temporally occur around the onset of microarousals, the occurrence of these electrophysiological events were compared in the two seconds before and eight seconds after the onset of any detected microarousal to their occurrence in time-matched control windows that were not associated with a microarousal (Fig. 4A). Only individuals who had at least five microarousals and five sleep spindles following microarousals were included (children = 29; early adolescents = 32; late adolescents = 27; young adults = 31) in these analyses. Temporal Relationship between Microarousals, Sleep Spindles, and High-Amplitude Slow Waves, and High-Amplitude Slow Wave-Sleep Spindle Coupling Independent of the Infraslow Fluctuation of Sigma Power. (A) Temporal distribution of high-amplitude slow waves (left) and sleep spindles (right) from two seconds before to eight seconds after microarousal onset. Event occurrence is expressed as the number of events in each 0.5 s time bin divided by the total number of events for that participant in the − 2 to + 8 s window, multiplied by 100, and then averaged across participants within each age group. Solid lines show events occurring around microarousals; dashed lines show events occurring in control windows outside the microarousal period, which provide a larger pool of control data. The standard deviation is represented by shaded areas. Asterisks above the curves indicate time bins with significant differences between around- and not-around-microarousal conditions. For visualization, group-averaged time courses are smoothed using a moving-average window applied to the plotted curves; statistical comparisons are performed on the underlying non-overlapping 0.5 s bins. (B) The number of spindles that couple into the up-phase of high-amplitude slow waves divided by the total number of spindles in the established time window, i.e., either in the 0.5 to two seconds after the onset of the microarousal (left) or any sleep spindle not occurring within this window (right; a larger pool of control data) is displayed. Each dot denotes a participant; participants are excluded from this analysis if they did not have at least five microarousals and five sleep spindles following the onset of microarousals. Significant differences are denoted with asterisk(s). Across all age groups, high-amplitude slow waves showed a significant increase in the 0.5 s immediately following microarousal onset (p-values < 0.05). In addition, the early adolescent, late adolescent, and young adult groups exhibited significant increases in the 0.5 s just before microarousal onset (p-values < 0.05). All age groups showed significant reductions in high-amplitude slow-wave occurrence in one or more bins within the five to eight seconds after the microarousal onset (p-values: <0.001–0.31). For sleep spindles, all age groups demonstrated significant reductions in one or more bins during the two seconds prior to the microarousal onset (p-values < 0.05), followed by significant increases in one or more bins from approximately 0.5 to two seconds after microarousal onset (p-values: <0.001–0.12). Additionally, the early adolescent, late adolescent, and young adult groups displayed significant reductions in some bins between six and eight seconds following the microarousal onset (p-values: <0.001–0.34). The sleep spindles that occurred from 0.5 to two seconds after the onset of microarousals were compared to all other sleep spindles in terms of coupling with high-amplitude slow waves (Fig. 4B). The percentage of sleep spindles coupling with high-amplitude slow waves was significantly higher for spindles occurring 0.5 to two seconds after the onset of microarousals when compared to all other spindles in late adolescent (z = 2.37; p = 0.02) and young adults (z = 2.76; p = 0.006). Discussion Our findings demonstrate that sigma power during N2 sleep fluctuates on an infraslow time scale from childhood to young adulthood, suggesting its fundamental biological and functional relevance. Yet, we observe that the frequency, variability, and strength of the ISFS increase from children and early adolescents to late adolescents and young adults. Notably, all age groups show a consistent organization in markers of arousal and memory reactivation, which occur the most in the spindle-rich peak of the ISFS and further support that the ISFS is intrinsic to sleep. To optimize and systematically capture age-specific aspects, we refined previous methods for analyzing the ISFS in humans2,8. These adaptations included extending bout lengths to cover at least two cycles of our lowest frequency of interest (0.0075 Hz), removing filtering and downsampling steps that may reduce the temporal resolution, utilizing FFT instead of wavelets for the spectral analysis, and revising criteria for detecting and quantifying the ISFS. Despite these methodological changes, the mean peak frequency detected in the young adult group (0.021 Hz) remained consistent with previously reported values (0.02 Hz) for this age range2, supporting the validity of our approach. Further analysis of the ISFS spectral parameters revealed significant developmental changes. Younger age groups exhibited lower peak frequency, indicating a longer average ISFS period, as well as narrower bandwidth, suggesting less variation in the ISFS period, and reduced area under the curve, reflecting less strength of the ISFS compared to older individuals. A recent nap study15 established age-specific normative values for sleep spindle characteristics, showing that spindle density increased from age three until about age 15, after which it plateaus into young adulthood. This aligns with our observation that ISFS parameters differ most significantly between younger (8 to 15 years old) and older (16 to 26 years old) groups. Given that the thalamus and cortex, the key regions involved in sleep spindle generation, undergo significant maturation during this period9,11,14, it is likely that these structural changes contribute to the developmental differences we observed in the ISFS parameters. However, the presence of the ISFS was observed across all age groups, and the investigation of topographical patterns revealed that the ISFS spectral parameters exhibit largely similar spatial distributions, suggesting the absence of a clear developmental gradient in the spatial organization and emphasizing the fundamental nature of this rhythm. Therefore, despite ongoing cortical and subcortical maturation of underlying systems11,38,39, the anatomical structures facilitating these infraslow dynamics appear to be established early in life. Only peak frequency was found to increase with age in a frontal cluster, which may reflect the prolonged maturation of the frontal cortex during childhood and adolescence30,40,41. The thalamocortical network generates and propagates sleep spindles, while the locus coeruleus modulates spindle dynamics on an infraslow timescale1,42,43. In rodents, locus coeruleus activity is both necessary and sufficient for generating the ISFS1. Analysis of the temporal dynamics of the ISFS has shown that when the ISFS is decreasing, norepinephrine is increasing, marking the fragile phase of the ISFS. Correspondingly, both around the peak of the ISFS and in the subsequent decreasing phase of the ISFS, external tones were more likely to induce wakefulness or microarousals (lasting four to 17 s and involving movement) in mice2,42,43. In contrast, decreasing locus coeruleus activity marks the protected phase of the ISFS, corresponding to around the trough and subsequent increasing phase of ISFS, which was associated with greater resilience to external stimuli and hippocampal ripple activity—an indicator of memory reactivation2,5,44. Expanding this to humans, we found that the onset of spontaneous microarousals (lasting three to 10 s and not necessarily involving movement) was not randomly distributed within the ISFS. Instead, they clustered around the peak of the ISFS in all age groups. Thus, the organization of arousal within the human ISFS shows some overlap with rodent findings, though it does not extend as far into the decreasing phase of the ISFS as observed in rodents. Consistent with this, a recent human study using pupil size as an index of locus coeruleus activity during sleep reported that pupil size fluctuates on an infraslow timescale, is time locked to the ISFS, and that its diameter increases (reflecting increased arousal) during the positive half-wave of the ISFS6. Likewise, human heart-rate dynamics show opposing relationships with the ISFS, further supporting a species-specific temporal shift likely driven by underlying anatomical or physiological differences, as previously speculated2,6,45. In addition, we observed that the percentage of high-amplitude slow waves and sleep spindles coupling with these slow waves, considered a marker of memory reactivation within the active-systems consolidation model46, was highest around the peak of the ISFS. This suggests that the temporal organization of both markers of arousal and memory follow a similar pattern within the human ISFS, though not aligned with findings from rodents. However, differences in how memory reactivation and microarousals are defined and measured across species, such as the microarousal duration, whether they are spontaneous or experimentally evoked, and whether muscle activity is required, combined with the absence of perturbation studies that directly manipulate ISFS dynamics in humans, hinder strong cross-species comparisons. At first glance, it may seem paradoxical that microarousals, brief events marked by increased cortical arousal, occur within the same phase of the ISFS as sleep spindles, which are traditionally considered sleep-protective47,48. Spindles are associated with higher sensory tolerance, and enhancing spindle activity increases the arousal threshold4,48–50. Yet (as well-described in a recent spindle review article4 in addition to being sleep-protective, spindles are also known to respond to perturbation. Namely, external stimuli can increase spindle density, typically alongside a high-amplitude slow wave (specifically, a K-complex), followed by a brief refractory period in which spindle activity is suppressed4,51–53. We find a similar pattern for microarousals: these spontaneous increases in arousal were often preceded or just followed by an increase in high-amplitude slow waves (as also found in54), which was then followed (0.5 to 2 s later) by an increase in sleep spindles, after which the occurrence of both event types decreased. Although speculative, this sequence, together with the clustering of microarousals, high-amplitude slow waves, and spindles within the same ISFS phase, may illustrate how these two spindle-related functions operate in a temporally organized manner within the ISFS. In this view, microarousals may mark brief fragile periods in which the sleeping brain samples the environment, and when no further action is required, spindle activity rapidly responds to stabilize the system and preserve sleep continuity, potentially in tandem with K-complexes as previously speculated47,55. Furthermore, this sequence may support a functional interplay between arousal and memory reactivation. The coordination of high-amplitude slow waves with sleep spindles is thought to facilitate hippocampo-cortical communication during memory reactivation, a process that may require short periods of increased cortical excitability to re-engage wake-active neuronal circuits46,56. This interpretation is supported by our finding that spindles occurring shortly after microarousal onset were more likely to be coupled to the up-phase of high-amplitude slow waves, particularly in late adolescents and young adults. Taken together, these findings suggest a coordinated interaction between transient arousal, thalamocortical synchronization, and memory reactivation within the ISFS during human NREM sleep. While functional markers show a clear organization within the ISFS that is maintained across development, spectral parameters of the ISFS and sleep spindles coupling to high-amplitude slow waves following microarousals show differences between childhood/early adolescence and late adolescence/young adulthood. Whether these differences contribute to increased difficulties with sleep initiation and maintenance, commonly reported in children and adolescent populations10,13, remains to be determined. Our study used retrospective datasets not specifically designed to perturb the ISFS or systematically assess overnight memory performance. Therefore, we only captured spontaneous arousal events, likely triggered by internal or random environmental factors, and memory reactivation inferred through electrophysiological markers. Further research exploring perturbations of the ISFS through external stimuli and its implications in various clinical populations and memory types will be crucial for understanding how these mechanisms shape sleep and memory throughout the lifespan. Moreover, the sex distribution across age groups was uneven, although follow-up analyses including sex as a covariate confirmed that age-related ISFS effects remained robust, aside from a modest influence on ISFS bandwidth. The distribution of electrophysiological markers within the ISFS was also qualitatively similar when analyzed separately in female and male participants. Finally, the ISFS was measured indirectly with surface EEG, making it difficult to infer both the precise temporal dynamics and underlying anatomical sources of the ISFS. In summary, our analyses demonstrate that sigma power fluctuates on an infraslow time scale across development, revealing age-specific differences in the ISFS spectral parameters. The increase in the frequency, variability, and strength of the ISFS across development may drive essential changes in the regulation of sleep and its relationship with memory reactivation. Markedly, we find that readouts of arousal and memory reactivation are organized within the spindle-rich peak of the ISFS irrespective of age group. By elucidating how key sleep processes are organized within the ISFS across development, this study lays the groundwork for future investigations into the role of ISFS in shaping cognitive development and arousal regulation, with implications for understanding both typical and disrupted sleep patterns from childhood to adulthood. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Acknowledgements We would like to thank the participants for their time and data contributions, as well as their families for their support. We also appreciate the invaluable advice and support from everyone in Reto Huber’s lab and Miriam Gerstenberg’s research group throughout this project. Author contributions Data Collection: SK, FP, VJ, CV, EK, MF, SL; Methodology: MED, AOF, SF, SV, PA, MG, RH; Formal Analysis: MED; Supervision: MG, RH; Writing - Original Draft: MED; Writing - Editing: MED, AOF, SF, SL, PA, MG, RH; Writing - Reviewing: MED, AOF, SF, SA, SK, FP, VJ, CV, EK, MF, SL, PA, MG, RH. Funding Sources Open access funding provided by University of Zurich. This work was supported by: UZH Candoc Grant (MED), Swiss National Science Foundation P500PB_214403 (AOF), Swiss National Science Foundation 320030_179443 (SF, MF, SL), Swiss National Science Foundation, 320030_153387 (CV, VJ, RH), HMZ Flagship grant “SleepLoop” (SL), Swiss National Science Foundation PCEFP1-181279 (SK), Frutiger Foundation (MG), EMDO Foundation (MG), Borbély-Hess Foundation (EK), Gottfried und Julia Bangerter-Rhyner-Stiftung (EK, RH). Data availability All data supporting the findings of this study are securely stored at servers of the University Children’s Hospital. Access and availability will be provided upon a material transfer agreement and after approval by the local ethics committee of the Canton of Zurich. Declarations Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Miriam Gerstenberg and Reto Huber contributed equally to this work. References Associated Data Supplementary Materials Supplementary Material 1 Data Availability Statement All data supporting the findings of this study are securely stored at servers of the University Children’s Hospital. Access and availability will be provided upon a material transfer agreement and after approval by the local ethics committee of the Canton of Zurich.