Background
Angelman syndrome (AS) is a neurodevelopmental disorder characterized by developmental delay, impaired speech and motor skills, and high comorbidity with epilepsy [
1]. Loss-of-function mutations in the maternal copy of the imprinted
UBE3A gene cause AS [
2,
3], while maternal duplications in the same region (15q11-13) are linked to autism [
4‐
6]. Recent work has identified multiple approaches with preclinical therapeutic potential for AS: antisense oligonucleotides and topoisomerase inhibitors have the potential to unsilence paternal
UBE3A and re-express UBE3A protein; gene therapy provides a direct method of expressing
UBE3A; mechanism-based approaches downstream of
UBE3A include GABA
A agonists (THIP/gaboxadol) and modulation of αCaMKII; other approaches include altering diet [
7‐
12]. Many of these approaches are in the pipeline for upcoming clinical trials. It is therefore critically important to develop biomarkers for AS that are clinically relevant, objectively quantifiable, highly penetrant, and have strong face validity between animal models and patient populations. Such biomarkers need not have predictive or diagnostic value, as AS diagnoses are confirmed genetically [
13], but rather their value would lie primarily in their use as outcome measures.
Electroencephalography (EEG) has revealed consistent signatures of AS, which have been described by clinical reports and case studies spanning nearly 30 years [
14‐
22]. EEG abnormalities in AS include rhythmic delta, rhythmic theta, and epileptiform spike-wave discharges. Increased delta rhythmicity is the most common EEG phenotype in AS (~84% of patients) [
21], and of these phenotypes, it is the most specific for AS relative to other syndromes [
20]. Multiple variants of delta activity have been described based on brain region and waveform characteristics [
20], yet every variant of delta, by definition, has a common oscillation frequency of ~2–4 cycles per second. Clinical studies typically report delta abnormalities in a binary fashion, being present or absent, but in some cases have further subdivided delta abnormalities into being continuous or intermittent [
17]. To date, no study has quantified delta rhythmicity in AS, quantitatively compared AS individuals to a neurotypical control group, or quantitatively tracked developmental and state-dependent (sleep/wake) changes in delta oscillations in AS. Principled characterization of these features, and validation in a mouse model, are critical for development of delta rhythms as a biomarker.
AS model mice (
Ube3a
m−/p+
) have genetic construct validity with the human condition and thus provide a powerful preclinical model. Silencing of the paternal
Ube3a allele is conserved from humans to mice; thus
Ube3a
m−/p+
mice, like individuals with AS, have minimal functional UBE3A protein [
23]. Using parallel quantitative methods, we analyzed delta rhythmicity in AS model mice and human EEG data. We found that increased delta power provides a robust and reliable biomarker with strong face validity between the AS mouse model and a patient population, 4–11 years old. Additionally, quantitative methods allowed for a novel study of delta “dynamics,” a measure of how delta rhythms vary over time across a single recording session. Delta activity is more dynamic, both in AS mice and AS individuals. Children with AS exhibited enhanced delta activity across all EEG electrode placements. The enhanced delta power and dynamics were present during both wakefulness and sleep and were observed at all ages tested but most pronounced in younger children. Overall, this study corroborates qualitative clinical descriptions of delta oscillations in AS individuals [
14‐
22], provides the first quantitative assessment of delta rhythmicity in AS individuals and comparison with a neurotypical reference group, and validates this biomarker in a mouse model. Delta rhythmicity thus has promise as a preclinical and clinical biomarker for AS and as an outcome measure for AS clinical trials.
Discussion
Rhythmic delta is the most pervasive EEG abnormality in AS, but delta phenotypes have not been previously quantified. If delta oscillations are to be an effective biomarker, quantitative methods are required to track acute or longitudinal changes in rhythmicity. Here, we used spectral analyses to confirm that delta abnormalities in AS model mice mirror clinical reports from the AS patient population (Fig.
1). Using similar methods, we quantified robust delta phenotypes in children with AS across the neocortex during wake and sleep (Figs.
3 and
4), showing that the enhanced delta phenotype scales in a state-dependent manner. The enhanced delta activity in AS individuals followed a predictable developmental trajectory across subjects and within subjects (Fig.
5). While delta phenotypes were stronger at earlier ages, they persisted in all age groups tested (4–11 years), demonstrating that delta activity may be useful as a longitudinal biomarker, in addition to its utility as an acute biomarker in young children. Spectral analyses revealed increased dynamics, or variability, of delta oscillations within single sessions in both AS model mice and children with AS (Figs.
2 and
4). This phenotype had not been described in a patient population and would be difficult to visualize and assess clinically without quantitative methods.
With multiple approaches currently being developed for clinical trials in AS, reliable and robust biomarkers are needed. Characteristics of a strong disease biomarker also include face validity and evidence for reversibility in a mouse model. Here, we showed that abnormal delta rhythmicity is conserved between mouse models and patient populations in AS, and prior work showed that increased delta power may be reversed in AS model mice by embryonic reinstatement of the UBE3A protein in a subset of neurons [
30]. To date, phenotypic behaviors have been characterized in AS model mice with varying reliability [
35] and include sensory, motor, and learning impairments [
11,
23]. Taken together, mouse behavioral phenotypes generally resemble human symptoms, but their direct face validity is limited and, thus, are not ideal biomarkers. One exception to this rule is seizures, which may be robustly and reliably induced in AS mouse models [
23,
30]. However, the use of seizures as a biomarker in AS children is limited; seizures are typically treated with antiepileptic medications and are controlled to a great extent in the majority of children [
36]. Delta rhythmicity represents a robust, reliable biomarker with strong face validity between mouse models and patient populations.
We observed strain differences in delta phenotypes in AS model mice: delta power (2–4 Hz) was increased in AS on a 129 background (Fig.
1), but not on a C57BL/6 background (Additional file
2: Figure S2), despite a trend towards increased power in the 3–5-Hz band. Despite statistically normal delta power, AS model mice on a C57BL/6 background did show increased delta dynamics (Fig.
2). Thus, while delta power phenotypes may be strain-specific, abnormal delta dynamics are preserved across two commonly used strains for AS research. Strain differences in delta power are not surprising, as behavioral differences have also been noted between AS mice on 129 and C57BL/6 backgrounds [
35].
Quantitative assessment of retrospective human EEG data revealed a robust increase in delta power in children with AS. These results support clinical reports [
14‐
22], and our data validate the utility of quantifying delta activity pre- and post-intervention to track acute and sustained consequences of therapeutic interventions. Spectral analyses also address the nature of delta abnormalities in AS in a manner not possible by clinician review. Our study of within-session delta dynamics revealed that delta oscillations are more variable in AS, but are not confined to intermittent bouts.
Clinically, delta abnormalities have been observed in both posterior (73% of patients) and anterior (59%) regions [
21], with potential differences in the type of delta seen by region [
37]. We found that delta phenotypes (increased power and dynamics) generalized across the neocortex in a large sample (Figs.
3 and
4). However, spatially restricted runs of delta were observed within individual recordings (Additional file
5: Figure S4). Additionally, while spectral analyses provide an unbiased method to quantify power within a band of interest, a disadvantage of their use is an inability to dissociate subtle variants of delta, such as notched delta (Additional file
5: Figure S4), which have been noted in clinical studies of AS [
17,
20,
22]. Thus, spectral analyses are best suited for quantifying broad delta biomarkers. We chose to focus on delta rhythmicity because it is the most common EEG abnormality in AS and the most specific abnormality to AS relative to related disorders [
20,
24]. However, interictal epileptiform discharges and theta abnormalities have also been widely reported [
14,
17,
20,
21,
38]. Epileptiform discharges are typically coincident with rhythmic delta [
37] and are therefore likely captured by using delta power as a biomarker; our analyses did not distinguish epileptiform discharges in the 2–4-Hz frequency band from background delta rhythms. Increased theta (~4–6 Hz in human) has been noted in ~30–60% of children with AS (Additional file
5: Figure S4), but is age-dependent and rarely observed beyond age 8 [
17,
20,
21]. Thus, we were not surprised to see normal theta in adult AS model mice (Fig.
1, Additional file
2: Figure S2). Quantitative assessment of theta and other bands in human EEG data were complicated by our a priori hypothesis that delta is increased and by the limits imposed by quantifying relative power (see “
Results” or “
Methods”).
Enhanced delta rhythmicity is a signature of slow-wave sleep, and our quantification confirmed that delta rhythms are indeed increased in neurotypical individuals during sleep epochs (Fig.
4). In AS individuals, delta rhythms are also increased during sleep relative to wakefulness, and thus, the enhanced delta phenotypes are preserved and scaled with state changes. These data show that it is critical to identify and separate wakeful and sleep epochs during EEG recordings but that delta remains an effective biomarker when making state-specific comparisons. Enhanced delta does not appear to broadly disrupt sleep architecture. Children with AS show typical sleep architecture such as sleep spindles and vertex waves. While there may be some disruption in sleep architecture, these appear to be minor compared to the significant effects of sleep-activated discharges on sleep architecture.
In addition to generalizing across sleep and wake, delta phenotypes in AS are also present across childhood development. We found a developmental reduction in delta power in AS; however, delta phenotypes persisted in all age ranges tested, to 12 years (Fig.
5). Thus, delta remains a valid biomarker throughout childhood and may be used as interventions and clinical trials are likely to occur in children of all ages. It is not clear whether the developmental attenuation of delta phenotypes is directly linked to loss of UBE3A. The attenuation of delta activity may be related to a secondary feature of AS, such as improvements in epilepsy and sleep at older ages [
39]. It is also not yet known how delta phenotypes correlate with clinical features of AS such as epilepsy severity, sleep, and behavioral, cognitive, and motor impairments. However, in mice, cell type-specific manipulations of UBE3A that increase delta power also increase seizure susceptibility, and those that do not affect delta also do not affect seizures [
30].
Our work represents the first direct comparison of EEGs from children with AS and neurotypical controls. However, an inherent limit of our retrospective EEG analyses was that AS data and neurotypical data were gathered at two different sites. We processed and analyzed all data in parallel and were encouraged by the robustness of phenotypes, but future prospective studies should be designed to recruit AS and control patients to a single site. In addition, intellectual disability in children with AS presents a potential confound, as EEG slowing has been associated with cognitive impairment in several populations [
40,
41]. Future work comparing AS to other reference groups (i.e., nonsyndromic seizure, intellectual disability, autism) will be critical to understanding the extent to which other disorders may exhibit delta phenotypes. AS may be considered an autism-like disorder, as a subset of children with AS also meet the diagnostic criteria for autism [
42‐
44]. Quantitative EEG methods have characterized some spectral and coherence phenotypes in nonsyndromic autism [
29,
45‐
51], yet the genetic heterogeneity of nonsyndromic autism introduces challenges in finding common EEG biomarkers. However, recent work has identified EEG signatures of Dup15q syndrome, a syndromic form of autism caused by duplication of the 15q11-13 genetic region which includes
UBE3A [
4‐
6]. The most profound EEG abnormality in Dup15q is increased beta rhythmicity, which is normal in AS model mice (Fig.
1, Additional files
1: Figure S1 and
2: Figure S2), but decreased delta power has also been noted in Dup15q individuals [
52‐
54]. Thus, bidirectional changes in
UBE3A gene dosage are linked to mirror symmetric changes in delta power, suggesting a critical role for UBE3A protein in regulating delta-generating brain circuits.
Fragile X syndrome, another single-gene disorder associated with autism, provides a case study in the importance of defining reliable biomarkers for use as clinical outcome measures. A series of mechanism-based pharmacological studies in mice sought to normalize synaptic protein synthesis, a key pathological feature of Fragile X [
55‐
57]. Pharmacological interventions directed towards normalizing protein synthesis were highly successful in correcting Fragile X phenotypes in mice [
58‐
60], ultimately leading to multiple phase 2 clinical trials [
61,
62]. These well-designed and well-powered trials ultimately failed because no improvements were seen in predefined behavioral endpoints [
63]. While other aspects of these studies were also relevant to their outcomes, such as the age of children enrolled and the duration of treatments, this work provides a rationale to develop biologically based, quantitative, robust, and repeatable outcome measures for clinical trials. We propose that delta rhythmicity meets these criteria for Angelman syndrome.
Acknowledgements
We thank April Levin (Boston Children’s), Rob Komorowski (MIT), Mark Shen (UNC), and Alana Campbell (UNC) for thoughtful discussion and advice. We thank Thorfinn Riday (Paris Descartes) for contribution of preliminary data and technical advice.