Background
Angelman syndrome (AS) is a neurodevelopmental disorder caused by loss of neuronal expression of the maternally inherited
UBE3A gene. Symptoms of AS include severe intellectual disability, impaired speech and motor function, epilepsy, sleep abnormalities, and some phenotypic overlap with autism [
1‐
3]. Consistent and widespread electroencephalographic (EEG) irregularities in AS include epileptiform discharges, intermittent theta waves, and enhanced rhythmic delta waves [
4‐
7]. In a prior study, we established that quantitative methods can be successfully applied to retrospective EEG data to confirm prior clinical descriptions of rhythmic delta in AS [
6]. Here, we sought to use quantitative approaches to identify novel EEG signatures in the same groups of retrospective EEG data. We assessed EEG coherence during wakefulness and non-rapid eye movement (NREM) sleep and quantified sleep spindles during NREM sleep.
Coherence is a measure of how two simultaneously recorded EEG signals are correlated and represents a non-invasive approach to assess functional connectivity between brain areas [
8]. We were motivated to study coherence in AS by the observation that individuals with autism show altered coherence patterns [
9‐
17]. Autism has been recognized as a component feature of AS [
18‐
22], and copy number increases in the 15q11-13 chromosomal region including
UBE3A are also associated with syndromic autism [
23,
24]. Some estimates suggest that up to ~ 50–80% of individuals with AS meet diagnostic criteria for autism [
18]; however, these estimates vary greatly due to the difficulties assessing autism with standardized clinical tests in AS individuals. Traditionally, individuals with autism were thought to have comparatively high coherence between nearby electrode pairs (local hyperconnectivity) and low coherence between long-distance signals (global hypoconnectivity) [
9‐
13], but this view has been challenged and become more nuanced in recent years [
14‐
17,
25]. Thus, although specific connectivity patterns remain unclear, there is widespread consensus that EEG coherence is altered in autism. The phenotypic and genetic links between AS and autism led us to hypothesize that children with AS might also display irregularities in the relationship between long-range and short-range coherence.
Sleep abnormalities are common in individuals with AS [
1‐
3,
26‐
34] and have also been reported in mouse models of the disorder [
35,
36]. Sleep dysfunction includes arousal during sleep and short sleep duration, and has a major impact on the quality of life of individuals with AS and their caretakers [
28‐
31]. We sought to identify quantitative EEG signatures underlying disrupted sleep patterns in children with AS. In addition to measuring coherence during sleep, we also quantified sleep spindles. Spindles are thalamocortical oscillations in the sigma band (~ 11–16 Hz) that occur during NREM sleep and are important for memory consolidation [
37,
38]. Sleep spindle activity is decreased in a number of neurodevelopmental and neurodegenerative disorders, such as autism, intellectual disability, epilepsy, Alzheimer’s disease, and schizophrenia [
39‐
46]. Although there have not yet been reports of substantial impairments in sleep architecture in AS, we hypothesized that quantitative measures might reveal subtle impairments in spindles and in patterns of sleep coherence that might be otherwise difficult to detect manually in a clinical EEG review setting.
During wakefulness, we report increased long-range EEG coherence in children with AS. During sleep, we also find increased long-range coherence, but specifically in the gamma band. We also report that sleep spindles are less frequent and shorter in children with AS. Overall, these measures provide insights into circuit-level neurobiology in AS and may have value as biomarkers or measures of target engagement for future therapeutic interventions. As this study was exploratory in nature, future work is needed to confirm coherence and spindle dysregulation in additional cohorts and to link these EEG phenotypes with behavioral outcomes.
Discussion
Quantitative EEG analyses revealed three phenotypes in children with AS that would otherwise be difficult to discern in a routine clinical or research setting: (1) increased long-range coherence during wakefulness, (2) increased long-range gamma-band coherence during sleep, and (3) decreased sleep spindle number and duration.
EEG coherence provides a measure of how neural activity is correlated between brain areas and is widely used as a proxy for functional connectivity [
8]. Coherence measures the consistency of the phase and amplitude difference between EEG signals in a given frequency band. Coherence is thus distinct from spectral power, which measures the relative amplitude of electrical activity within a frequency band from a single electrode. Thus, despite robust increases in delta power [
5,
6], children with AS have normal delta-band coherence (Figs.
2 and
3). While coherence and delta power phenotypes in AS are both ultimately caused by loss of neuronal UBE3A protein, they likely reflect different proximate circuit-level impairments.
During wakefulness, long-range EEG coherence was increased in children with AS across a broad range of frequencies (Fig.
2). Increased long-range coherence in AS was seen throughout the brain and was not driven by altered coherence in a spatially restricted subset of connections (Fig.
2e, Additional file
1: Figure S1). There is general consensus that functional connectivity is widely disrupted in autism [
9‐
17,
25], and our findings confirm that coherence is also dysregulated in AS, a disorder with some autistic features. However, increased long-range functional connectivity may be surprising given prior studies of decreased structural connectivity in AS, both in mouse models [
65] and patient populations [
66,
67]. This suggests that despite reduced structural connectivity, there may be fewer inhibitory constraints on efferent projections in the AS brain.
During sleep, long-range coherence was significantly increased in children with AS, but only in the gamma band (Fig.
3). Gamma-band coherence is an indicator of attentive wakefulness [
68], and accordingly, gamma coherence is typically lower during sleep than during wakefulness [
69‐
71]. We confirmed that gamma coherence in neurotypical children is lower during sleep than during wake (compare Figs.
2 and
3). However, the pattern of elevated long-range gamma coherence during sleep in AS children resembles what is typically seen in a wakeful state. A common challenge in analyzing gamma-band coherence is the presence of electromyogenic artifacts, which are visible in EMG spectra and are often seen temporally in the gamma range [
53,
54]. Therefore, we used an outlier analysis to exclude recordings in which EMG artifacts exceeded an established threshold [
56]. Two additional pieces of evidence confirm that gamma coherence phenotypes in AS are not driven by EMG artifacts: (1) increased gamma coherence is specific to long-range electrode pairs and (2) gamma coherence is not increased specifically in temporal electrodes (Additional file
2: Figure S2I). Overall, long-range functional connectivity was increased in AS EEGs during both wake and sleep states. However, coherence patterns differed as function of state: phenotypes were gamma-specific during sleep and not frequency-specific during wake. Thus, it is critical to control for sleep state when assessing functional connectivity.
We also report that sleep spindles are shorter and less frequent in AS (Fig.
4). This finding is consistent with the decreased spindle frequency seen in autism, intellectual disability, and sleep disorders [
39‐
43]. Despite many clinical studies of Angelman EEGs over the past 30 years, to our knowledge, there have been no reports to date of dysregulated spindles. This is surprising because unlike coherence, sleep spindles may be easily detected by the eye. However, subtle dysregulation of spindles may be difficult to gauge clinically, especially given the pervasive disruptions in background activity in AS [
5]. Therefore, automated spindle detection using an unbiased, high-throughput method was used to determine that spindle rate and duration were decreased in AS EEGs. In addition, one of two blinded experts confirmed a statistically significant decrease in spindle rate in AS EEGs, with the other finding a strong trend. To enable blinded data analysis, we filtered out the delta activity that is pervasive in the AS EEG; however, this likely reduced both accuracy and confidence of manual detection. Future studies of sleep spindles in AS must consider and weigh the challenges of manual and automated detection, but we favor an automated approach because it is not subject to the reporter biases that plague qualitative outcome measurements in clinical trials.
More broadly, experimental conditions must be considered when evaluating our exploratory analyses of sleep composition in AS (both spindles and coherence). We used retrospective EEG data, which included periods of sleep and wake and was not designed explicitly as a sleep study. Because children with AS have pervasive sleep problems, it is likely that sleep quality during EEG recordings varied by group. For example, only 46% (13/28) children with AS slept during EEGs, whereas 75% (54/72) of neurotypical children slept. In addition, the nature of sleep during clinical EEG recordings may not be representative of typical overnight sleep. For example, the average length of NREM sleep during EEGs recordings was only ~ 14 min for neurotypical children and ~ 22 min for children with AS [
6]. Thus we propose that sleep spindles and gamma coherence phenotypes should be explicitly tested in well-controlled overnight sleep studies.
Clinical trials are on the horizon for AS; therefore, development of biomarkers, outcome measures, and measures of target engagement are especially valuable. Biomarkers for AS need not have diagnostic value, as diagnoses are made genetically. Therefore major considerations in evaluating a biomarker include whether it is quantitative, easily measured, reliable, and linked to clinically meaningful outcomes [
64]. Previously, we described enhanced delta rhythmicity in AS, which is quantitative, non-invasive, and reliable, but the link between delta rhythms and behavior has not yet been established. While effect sizes of gamma coherence and sleep spindle phenotypes are less than delta rhythms (Table
1), these phenotypes are likely linked to sleep quality. Therefore, they may be considered as biomarkers, particularly if a study is interested in quantifying sleep as a primary outcome measure. However, delta power is a substantially more robust biomarker, with only slight overlap between AS and neurotypical groups at the level of individuals. Future study of sleep biomarkers in an overnight setting, with AS and neurotypical children studied in parallel at a single site, may have the potential to decrease individual variability and increase robustness.
Quantitative EEG phenotypes may also provide insights into circuit-level biological mechanisms underlying AS. For example, mechanisms governing spindle initiation and propagation have been well characterized [
37]. Spindles are driven by the intrinsic properties of, and interactions between, thalamocortical cells and thalamic reticular cells. Thalamocortical circuits, which also drive cortical delta rhythms [
72], may be studied in mouse models to better understand how loss of UBE3A disrupts neural circuits. We hypothesize that loss of UBE3A from a small population of like neurons is sufficient to disrupt sleep spindles in AS. Coherence phenotypes, which are expressed broadly throughout the brain, are likely driven through different processes.