Editor's evaluation: Non-rapid eye movement sleep and wake neurophysiology in schizophrenia

Abstract

Article Figures and data Abstract Editor's evaluation Introduction Results Discussion Materials and methods Data availability References Decision letter Author response Article and author information Metrics Abstract Motivated by the potential of objective neurophysiological markers to index thalamocortical function in patients with severe psychiatric illnesses, we comprehensively characterized key non-rapid eye movement (NREM) sleep parameters across multiple domains, their interdependencies, and their relationship to waking event-related potentials and symptom severity. In 72 schizophrenia (SCZ) patients and 58 controls, we confirmed a marked reduction in sleep spindle density in SCZ and extended these findings to show that fast and slow spindle properties were largely uncorrelated. We also describe a novel measure of slow oscillation and spindle interaction that was attenuated in SCZ. The main sleep findings were replicated in a demographically distinct sample, and a joint model, based on multiple NREM components, statistically predicted disease status in the replication cohort. Although also altered in patients, auditory event-related potentials elicited during wake were unrelated to NREM metrics. Consistent with a growing literature implicating thalamocortical dysfunction in SCZ, our characterization identifies independent NREM and wake EEG biomarkers that may index distinct aspects of SCZ pathophysiology and point to multiple neural mechanisms underlying disease heterogeneity. This study lays the groundwork for evaluating these neurophysiological markers, individually or in combination, to guide efforts at treatment and prevention as well as identifying individuals most likely to benefit from specific interventions. Editor's evaluation This study, one of the largest of its kind, replicates previous findings regarding the impairment of sleep rhythms in patients with schizophrenia relative to healthy controls. Specifically, sleep spindles, which constitute a hallmark of non-Rapid Eye Movement sleep, are less frequent in people with schizophrenia and several other sleep features were also affected. Overall, this study provides evidence that brain dynamics during sleep are promising biomarkers for the diagnosis and the prevention of schizophrenia. https://doi.org/10.7554/eLife.76211.sa0 Decision letter Reviews on Sciety eLife's review process Introduction Schizophrenia (SCZ) is a chronic disorder characterized by cognitive and behavioral dysfunction that significantly impacts the quality of life of affected individuals and their caregivers (Charlson et al., 2018; Stanley et al., 2017). It is highly heritable, exhibiting a heterogenous and polygenic architecture implicating many genetic risk factors (Ripke, 2014). Despite intensive research, current medications ameliorate positive symptoms only in a subset of individuals, often with side effects and with little impact on negative symptoms or cognitive deficits. Consequently, recovery outcomes have not improved over the past decades (Jääskeläinen et al., 2013). Given the clinical and genetic heterogeneity of SCZ, identifying reliable, objective biomarkers that index specific neurobiological deficits is crucial for developing the next-generation therapeutics. Emerging evidence points to thalamus as a critical node that supports cognitive function, and to abnormal thalamocortical connectivity as a key neurobiological deficit in SCZ (Woodward et al., 2012; Anticevic et al., 2015; Cao et al., 2018; Woodward and Heckers, 2016; Wu et al., 2022). NREM (non-rapid eye movement) sleep offers a lens through which we may index thalamocortical function without confounds from waking behaviors such as active symptoms or altered motivation. Two hallmarks of NREM sleep measured by the electroencephalogram (EEG) – slow oscillations (SO) and spindles – reflect distinct thalamic and thalamocortical circuits. The slow (~1 Hz) neuronal oscillations with large amplitude are generated by cortical neurons and propagated by cortico-thalamo-cortical circuits, while spindles are bursts of oscillatory neural activity (typically between 10 and 16 Hz and ~1 s in duration) arising from reverberant interaction between thalamic reticular nucleus (TRN) and thalamocortical relay neurons and modulated by thalamocortical connections. The coupling of SOs and spindles mediates information transfer and storage during sleep that underpins NREM’s role in overnight memory consolidation (Walker and Stickgold, 2004). NREM traits show strong heritability in healthy populations (Ambrosius et al., 2008), correlate with cognitive performances (Geiger et al., 2011; De Gennaro et al., 2008; Purcell et al., 2017), and afford objective, quantifiable markers of thalamic and thalamocortical functioning in large cohorts. Impaired sleep spindles and SOs have been reported in SCZ patients. Despite relatively consistent support for a general reduction in spindle density and/or amplitude in SCZ patients (Ferrarelli et al., 2007; Ferrarelli et al., 2010; Göder et al., 2015; Mylonas et al., 2020a; Schilling et al., 2017; Wamsley et al., 2012), it is not clear how specific topographical and temporal spindle properties, which may reflect distinct thalamocortical connections, are associated with SCZ. For example, distinguishing fast and slow spindles (FS and SS) and their temporal coupling with SO are less studied in SCZ patients. More generally, a comprehensive examination of how NREM features (including existing and novel metrics) are altered in SCZ and how they relate to each other is not established (Castelnovo et al., 2018; Zhang et al., 2020). This is in part due to the relatively small sample sizes (N < 30) previously utilized, that were unable to support comprehensive analyses across multiple domains of NREM sleep. More importantly, whether and how distinct NREM sleep deficits track with waking EEG, clinical symptoms, medication, and cognition within patients is not clear (Au and Harvey, 2020). For example, one recent report suggested reductions in spindle density were significantly more extensive in patients experiencing auditory hallucinations compared to hallucination-free patients (Sun et al., 2021). Such a link is intriguing, considering the substantial evidence of altered auditory processing in SCZ (Erickson et al., 2016; Freedman et al., 2020; Thuné et al., 2016), and that both spindles and auditory processing heavily rely on uninterrupted and precise function of thalamocortical circuits (Ferrarelli and Tononi, 2010). However, whether auditory abnormalities during wake and spindle deficits during sleep reflect similar or distinct thalamocortical dysfunctions in SCZ are yet to be investigated. Here, using whole-night high-density electroencephalography (hdEEG), we comprehensively characterized alterations across multiple domains of NREM neurophysiology, their interdependencies and relationship to wake and clinical features in a new SCZ cohort of 72 patients and 58 healthy controls. We first sought to replicate the primary spindle density deficit and characterize the precise facets of spindle activity associated with SCZ, including topography, morphology, and novel metrics of intra-spindle frequency modulation. Second, we extend our analyses to broader NREM sleep EEG, including spectral features, functional connectivity, SOs, and their coupling with spindles, to determine which features track with disease state, and whether their association with SCZ is independent of spindle parameters. Third, we asked whether spindle deficits likely index the same thalamocortical dysfunctions reflected in auditory processing abnormalities during wake, by determining how sleep and wake metrics were correlated within individuals. Fourth, we investigated the relationship between disease-associated metrics and symptom severity within patients. Fifth, we attempted to replicate our sleep EEG findings by compiling an independent and demographically distinct replication dataset of 57 cases and 59 controls and applying the same analytic procedures. Finally, based on multiple domains of EEG metric, we constructed joint models to classify diagnostic status and assessed their predictive performance in the original sample, as well as their transferability to the independent collection of patients with distinct demographics. Results Whole-night sleep EEG and wake auditory event-related potential (ERP) data were acquired from 130 individuals – 72 patients diagnosed with SCZ (25 females, 34.8 ± 7.0 years of age) and 58 healthy volunteers (CTR, 25 females, 31.7 ± 6.3 years of age) with no personal or family history of SCZ spectrum disorders. As SCZ patients were on average older than controls (effect size [e.s.] = 0.48 standard deviation (SD) units, p = 0.014), all statistical comparisons were performed using logistic regression with age and sex as covariates (see Materials and methods for details). Sleep stage macro-architecture is largely unaltered in SCZ Based on manual sleep staging in 30 s epochs (see Materials and methods), the duration and proportion of sleep stages did not differ between SCZ and CTR groups, with the exception of the proportion of N2 sleep relative to total sleep time (TST), which was smaller in SCZ (e.s. = –0.38 SD, p = 0.048). For the primary analytic sample, we removed four subjects (due to persistent line noise artifacts, see Materials and methods for details); in this final N = 126 sample, there were no significant case/control differences in stage duration (p = 0.14 for N2 percentage). Patients spent longer time in bed (TIB) before sleep onset (sleep latency, 50 versus 24 min, e.s. = 0.95 SD, p = 0.0003), with concomitant differences in TIB (e.s. = 1.47 SD, p = 2×10–7) and sleep efficiency (e.s. = −0.8 SD, p = 0.0013). Wake after sleep onset (WASO) time was not significantly different between groups (p = 0.13), suggesting that reduced sleep efficiency in SCZ was attributed to later sleep onset rather than more fragmented sleep. As a confirmation, group differences in sleep efficiency were not significant after controlling for sleep latency (p = 0.1). Congruent with the sleep EEG night, data from a sleep habit questionnaire, a 2-week sleep journal, and information on sleep and wake times the night before the EEG recording all indicated that SCZ patients had earlier bedtimes (all p < 10–10) and longer TIB (all p < 10–8), suggesting that the EEG night followed a typical sleep/wake schedule for SCZ patients. N2 sigma power is decreased in SCZ Sleep EEG analyses were based on 57 channels, resampled to 200 Hz and re-referenced to the linked mastoids. We extracted all N2 epochs and applied automated artifact detection/correction; all sleep EEG analyses used our open-source Luna package (see Materials and methods). With the exclusion of four patients due to persistent line noise artifacts, the final sleep EEG dataset comprised of 68 SCZ and 58 CTR, with an average of 353 ± 143 and 374 ± 89 N2 epochs, respectively (no group difference, p = 0.16). Replicating prior studies (Ferrarelli et al., 2007; Ferrarelli et al., 2010; Manoach et al., 2010; Manoach et al., 2014), sigma-band (11–15 Hz) spectral power, a common proxy for spindle activity, was significantly reduced in SCZ (for power differences in classical frequency bands, see Figure 1—figure supplement 1). Figure 1 (top row) illustrates spectral power from 0.5 to 20 Hz averaged over Fz, Cz, and Pz channels, with shaded regions indicating significant group differences (p < 0.01, unadjusted for multiple comparisons). Across all channels, the largest sigma-band deficit (–0.88 SD) was observed for FC1 at 13.25 Hz. Adjusted for multiple comparisons (see Materials and methods), 34 channels showed deficits (padj < 0.05) at 13.25 Hz (Figure 1, bottom right). Power in delta and lower theta frequencies (~2–6 Hz) was also reduced in SCZ (Figure 1, top row) with a maximum effect size at 3.25 Hz for Cz (e.s. = –1.21 SD) and 46 channels showing significantly (padj < 0.05) lower power in SCZ (Figure 1, bottom left). Figure 1 with 1 supplement see all Download asset Open asset Decreased power in 2–6 Hz and spindle frequencies in schizophrenia (SCZ) during N2 sleep. The top row illustrates spectral power for frequencies from 0.5 to 20 Hz at Fz, Cz, and Pz channels averaged for SCZ (purple) and CTR (green) groups. Dotted black line illustrates the difference (SCZ-CTR) in averaged spectral power (10×Δlog10(PSD)) and frequency ranges marked with vertical gray bars illustrate frequency bins that showed significant group differences (p < 0.01, unadjusted). The plots in the bottom row illustrate power spectral density (PSD) averaged within groups and spatial distribution of group differences across channels. Channels with p-values adjusted for multiple comparisons (padj) < 0.05 are encircled with black line. The color of each channel corresponds to the signed −log10 adjusted p-value indicating the direction of group differences (blue corresponds to reduction, and red corresponds to increase in SCZ). Two scatterplots show individual spectral power for SCZ and CTR groups for channels with the largest effect sizes for the two indicated frequencies (marked with a white cross on the topoplots). Spindle density is reduced in SCZ, accompanied by altered spindle morphology We detected discrete spindle events using our previously reported wavelet-based algorithm (Purcell et al., 2017), targeting SS and FS separately by setting wavelet center frequencies to 11 and 15 Hz, respectively. Spindle densities showed the expected topographies, being maximal at either central/parietal (FS) or frontal (SS) channels (Figure 2). FS density was globally decreased in SCZ, significant at padj < 0.05 for 53 of 57 channels, with the largest deficit (32% reduction and e.s. of –1.27 SD) at FC2 (nominal p = 4×10–6). At this channel, average FS density in the CTR group was 2.7±0.7 spindles per minute, compared to 1.9±0.8 in the SCZ group. SS also exhibited reduced densities, albeit mostly restricted to posterior channels (max e.s. –0.79 SD at P7, reduction = 28%, nominal p = 5×10–5). Taken together, these analyses confirmed reductions in both sigma power and spindle density with high statistical confidence, providing a clear and independent replication of prior reports. Figure 2 with 3 supplements see all Download asset Open asset Reduced density of slow (SS) and fast spindles (FS) during N2 sleep in schizophrenia (SCZ) patients observed together with alterations in spindle morphology. The top two rows show topographical distribution in density, amplitude, integrated spindle activity (ISA), duration, chirp, and frequency of SS and FS averaged across SCZ (first row) and CTR (second row) groups. The third row of topographical plots illustrates group differences in these metrics. Each circle represents an EEG channel. EEG channels with adjusted p-values (multiple comparisons) < 0.05 are encircled in black and the color of each channel corresponds to the signed −log10 adjusted p-value indicating the direction of group differences (blue corresponds to reduction, and red corresponds to increase in SCZ). The bottom row illustrates distributions of spindle parameters in SCZ and CTR groups in the channel with the largest effect size of group differences (marked with a white cross on the topoplot). For each spindle, we further estimated the amplitude (maximal peak-to-peak voltage), duration, integrated spindle activity (ISA, a normalized measure that reflects both duration and amplitude), frequency, and chirp (intra-spindle change in frequency, indexing the typical deceleration of a spindle oscillation over its course) (Figure 2, Supplementary file 1). See Materials and methods for details. For SS, we observed an extensive SCZ-associated reduction in amplitude (50 channels with padj < 0.05, max e.s. –0.82 SD with 18% reduction at CP3) and ISA (33 channels with padj < 0.05, max e.s. –0.96 SD at CPZ). Reductions for FS were topographically more limited for amplitude (23 central/parietal channels with padj < 0.05, max e.s. –0.62 SD with 15% reduction at Cz) and ISA (single channel, F1, padj = 0.037, e.s. = −0.73 SD). FS (but not SS) were shorter in duration for SCZ in all but prefrontal and temporal channels (36 channels with padj < 0.05, max e.s. –0.98 SD at P5). In addition, FS chirp was more negative in SCZ (20 posterior channels with padj < 0.05, max e.s. –1.09 SD at O1), indicating more prominent deceleration in SCZ, that is, stronger chirp. Both SS and FS showed comparable SCZ and CTR distributions of observed average spindle frequencies, however. Densities of SS and FS were not significantly correlated with each other (abs(r) < 0.15 and p > 0.05 for SS and FS at Cz or Fz, as well as SS at Fz compared to FS at Cz). With respect to density reduction in SCZ, SS and FS effects were also statistically independent. For example, group differences in SS density at P7 (largest e.s. of the group differences, p = 5×10–5) were still significant (p = 6×10–4) when FS density (at P7 and FC2 – largest e.s. of differences in FS density) was added as a covariate in a joint model. The same was true for FS density at FC2 (original p = 4×10–6 and after controlling for SS density, p = 4×10–5). Below (see section Dimension reduction to summarize sleep EEG alterations), we explore in more detail the correlational structure of sleep EEG metrics, quantifying the underlying, independent components of variation across spectral, spindle, and other metrics, accounting for dependencies across scalp topography and neighboring frequencies. We confirmed that observed spindle density alterations extended to entire NREM sleep (N2 and N3 combined, data not shown) but expressed a sleep cycle-dependent effect for both SS and FS with more extensive group differences in the later sleep (cycle 3 compared to cycle 1, Figure 2—figure supplement 1). Sensitivity analyses to address potential medication effects Since 67 of 68 patients in the analytic sample were taking antipsychotic medication (Supplementary file 2), we converted antipsychotic doses to chlorpromazine equivalents (Langan et al., 2012), to determine whether group differences in spindle characteristics were likely to reflect medication effects (Supplementary file 2). There were no significant (all unadjusted p > 0.01) associations between total antipsychotic dose and any of the spindle parameters altered in SCZ, with the exception of FS chirp. However, this latter effect, seen only at two channels, had a direction of effect opposite to group differences observed in SCZ (FC6, t-value = −2.83, p = 0.006 and TP8, t-value = 2.8, p = 0.007), meaning that the original FS chirp SCZ association did not simply reflect antipsychotic dose. Further, although multiple spindle parameters correlated with clozapine usage (N = 12 cases, Supplementary file 2), all original SCZ-CTR differences remained significant after patients using clozapine were removed from the analyses (Supplementary file 3). With respect to adjunctive medication such as sedatives and tranquilizers, mood stabilizers and antiepileptics, and anticholinergics (Supplementary file 3, top 8 rows), group differences in spindle metrics persisted after controlling for each class, except for FS ISA. Taken together, these results strongly suggest that the observed group differences cannot be directly explained by medication effects, consistent with prior studies showing spindle deficits in unmedicated patients and first-degree relatives (Manoach et al., 2014). SO abnormalities in SCZ We detected SOs in N2 sleep using a previously described heuristic, based on relative/adaptive amplitude thresholds (see Materials and methods). Patients had an increased density of SOs across posterior channels (30 channels with padj < 0.05, max e.s. = 1.16 SD at O1, 18% increase), longer SO duration/wavelength (36 channels with padj < 0.05, max e.s. 1.92 SD at Cz, 18% increase) and flattened SO slope (44 channels with padj < 0.05, max e.s. –1.14 SD at Cz, 24% decrease) compared to controls (Figure 3, left panel). No differences between groups were observed in negative peak or peak-to-peak amplitude. Figure 3 with 1 supplement see all Download asset Open asset Altered slow oscillations (SOs) and their coupling with spindles in schizophrenia (SCZ). Left: Rows represent the topographical distribution of different SO parameters averaged across SCZ (first row) and CTR (second row) groups. Right: Coupling between SO and spindles in SCZ and CTR groups. Each point represents coupling magnitude and average SO phase of a spindle peak at a given channel in each participant. Mean SO angle of coupling for slow and fast spindles is shown as a vertical line of the corresponding color. The black line illustrates SO waveform at corresponding phase angles. Topoplots in the third row illustrate significance and direction of group differences across electroencephalogram (EEG) channels in SO parameters and its coupling with spindles. EEG channels with significant group differences (padj < 0.05) are encircled in black and the color of each channel represent the signed −log10 adjusted p-value indicating decrease (shades of blue) or increase (shades of red) in the tested EEG parameter in SCZ compared to CTR. Individual data at channels with the largest effect sizes of group differences are provided in the bottom row of scatter plots for the altered EEG parameters in SCZ (the exemplary channel is marked with a white cross on the topoplot above). SO parameters were not associated with total dose of antipsychotics but were affected by adjunctive medications: namely, increased SO duration and decreased slope were significantly associated with use of sedatives and tranquilizers; reduced SO slope was also linked to mood stabilizing and antiepileptic medication (uncorrected p < 0.01, Supplementary file 2). However, all SCZ-CTR group differences persisted in sensitivity analyses controlling for medication dosage (Supplementary file 3). Of note, group differences in SO characteristics depended on the use of absolute versus relative amplitude thresholds for detecting SOs, as we have noted in other contexts (Djonlagic et al., 2021). We adopted relative amplitude thresholds for our primary analyses, as it maximized the magnitude of spindle/SO coupling observed across the whole sample. If using an absolute threshold, differences in SO density, duration, and slope were attenuated, whereas SO negative peak amplitude achieved significant group differences. In addition, as SOs are most characteristic of N3 sleep, we performed secondary analysis that combined N2 and N3 sleep periods. While alterations in duration and slope of SOs remained significantly different between the groups, there were no differences in SO density applying either absolute or adaptive amplitude threshold for SO detection. A cycle specific analysis of N2 using adaptive threshold revealed that SO density alterations were more pronounced later in the night (Figure 2—figure supplement 1). Preserved coupling strength but altered overlap between SS and SO We quantified SO/spindle coupling in three ways, using empirically derived surrogate distributions to control for chance coupling (see Materials and methods): the rate of gross overlap between spindles and SOs, the magnitude of coupling based on the non-uniformity of the distribution of SO phase at spindle peaks, and the mean phase angle derived from that distribution. As expected, we observed a marked, non-random temporal coupling between both FS and SS and SO phase (Figure 3, top right). Consistent with previous reports (Djonlagic et al., 2021) whereas FS peaks (points of maximal peak-to-peak amplitude) were enriched on the rising slope of the SO and near the SO positive peak (mean phase = 240.8 degrees in CTR), SS tended to peak later (mean phase = 13.8 degrees in CTR) (Figure 3, top right). In CTRs, 97% of subjects had significant SO phase coupling for FS at Cz (empirical p-value < 0.05 obtained by randomly shuffling spindle peak timing with respect to SO phase to generate the null distribution) and 76% had significant SS coupling at Fz. The SCZ group displayed broadly comparable proportions (94% and 68% of cases showing evidence of non-uniform coupling, for FS at Cz and SS at Fz, respectively). We did not observe any significant group differences in the average magnitude of SO phase/spindle coupling based on the surrogate distribution normalized metric (Figure 3, right panel; see Figure 3—figure supplement 1 for the topographical distribution of coupling characteristics averaged across SCZ and CTR groups). With respect to SO phase, both SS and FS tended to occur earlier in SCZ, albeit this effect was only observed at a few parietal channels therefore precluding any strong conclusions (P5 and P7 for SS, max e.s. = −1.07 SD and P4, P6 for FS, max e.s. = −0.65 SD). The most pronounced coupling differences were in gross overlap of SS (the number of spindles showing any overlap with an SO, controlling for spindle and SO density), which was decreased in SCZ (13 channels with padj < 0.05, max e.s. –1.14 SD at F3). None of the coupling metrics were associated with total dose of antipsychotics but there was increased SS-SO overlap in patients taking aripiprazole or clozapine; an opposite effect was observed for mood stabilizing and antiepileptic medication (Supplementary file 2). Intra-spindle frequency modulation: chirp/deceleration and SO phase As noted, we observed a greater deceleration of FS (chirp) in SCZ. This effect was statistically independent of the reduction in FS density: for example, in a joint model, FS chirp and density at Pz were both independent predictors of SCZ (FS density t-score = –3.77, p = 0.0002 and chirp t-score = –3.17, p = 0.0015). Given that spindles are temporally coupled with SO phase, as well as the fact that we observed SCZ/CTR differences for both FS chirp and preferential SO phase at spindle peak, we asked whether spindle deceleration was dependent on SO phase. We summarized spindle instantaneous frequency (estimated using a filter-Hilbert approach, see Materials and methods) as a function of spindle progression (quantiles of spindle duration) and also SO phase (in eighteen 20-degree bins). Considering first only spindle progression, we observed the characteristic negative chirp (deceleration during the later portion of the spindle), for both FS and SS across all channels (Figure 4, top row). We further observed that spindle frequency was modulated by SO phase, consistently across channels but differently for FS and SS, likely reflecting the different temporal SO phase coupling of FS and SS (Figure 4, middle row): for SS, the highest instantaneous frequencies coincided with the SO positive peak, but earlier for FS (Figure 3, top right). Figure 4 with 1 supplement see all Download asset Open asset Intra-spindle frequency deceleration (chirp) is associated with slow oscillation (SO) phase differently for slow and fast spindles and such phase-frequency modulation is attenuated in schizophrenia (SCZ). Top: frequency changes across slow and fast spindle as spindles progress in five quintiles across all channels (averaged across all participants). Middle: intra-spindle frequency dependency on SO phase across all channels (lines) and total number of spindles detected in all channels and individuals at a given SO phase bin (gray bars). Bottom: spindle frequency as a function of both spindle progression (y axis) and SO phase (color of the curve; each curve represents frequency of spindles averaged across all channels for each 20-degree phase bin; blue curves illustrate frequency of spindles occurring close to the negative peak and red curves illustrate spindles next to the positive peak of SOs). The topoplots on the right illustrate topographical distribution of phase/frequency modulation averaged within SCZ and CTR groups separately for slow and fast spindles. Phase/frequency was estimated as linear-circular correlation between instantaneous frequency of individual spindle and phase of co-occurring SO. Topoplots in bottom right illustrate group differences in phase/frequency modulation. Channels with uncorrected p-values < 0.01 are encircled in black, and the color of each channel represents the signed −log10 p-value indicating decreased (shades of blue) or increased (shades of red) coupling in the SCZ group. Spindle progression and SO phase were related, reflecting SO/spindle coupling. We therefore additionally summarized spindle instantaneous frequency as a joint function of both spindle progression and SO phase, that is, 5 spindle progression quantiles x 18 SO-phase bins = 90 combinations averaged over all channels (Figure 4, lower row, Figure 4—figure supplement 1). This analysis suggested that spindle progression and SO phase have independent, largely additive effects on spindle frequency, as each line (SO phase bin) shows broadly similar chirp. Likewise, conditional on spindle progression, the SO modulation effect is still evident by the vertical offsets of the different lines (SO phase bins). More formally, modeling SS instantaneous frequency as a cubic function of spindle progression and/or SO phase (CTRs only) showed an associated adjusted R2 = 0.48 and 0.44, respectively, and R2 = 0.95 if combined. For FS, SO phase explained relatively less of the variance in spindle frequency (R2 = 0.19 versus R2 = 0.72 for spindle progression; combined, R2 = 0.94). That is, SO phase has a modulating effect on spindle frequency, independent of the typical deceleration/chirp, and this modulation was stronger for SS than FS. To distinguish this from temporal spindle/SO coupling, we refer to it as ‘phase/