Background
Autism is a common neurodevelopmental disorder affecting 1 in 110 children [
1], but its genetic etiology is complex and heterogeneous [
2,
3] Among the leading genetic causes of autism are abnormalities in proximal chromosome 15q, collectively referred to as "duplication 15 syndrome" (dup15q), which occur in ~1-3% of autism cases [
4‐
6]. Dup15q syndrome is a clinically heterogeneous neurodevelopmental disorder characterized by varying degrees of cognitive impairment, gross motor delays, seizures, dysmorphic features and autism in 85% of cases [
5,
7,
8].
Chromosome 15q11-q13 is a genetic hotbed for neurodevelopmental disorders because of a high density of low copy repeats (LCRs) that increase susceptibility to errors in meiotic recombination, resulting in deletions as well as duplications [
9,
10]. The parental origin of deletions and duplications influences the expression of 15q11-q13 through genomic imprinting. Imprinting of 15q11-q13 is regulated by the bipartite imprinting center (IC), which contains a differentially methylated region at the 5' end of
SNRPN that controls expression throughout 15q11-q13 [
11]. Loss of 15q11-q13 paternally expressed genes through deletions or maternal uniparental disomy (UPD) results in Prader-Willi syndrome (PWS), whereas the maternal deficiency at this locus results in a phenotypically distinct neurodevelopmental disorder, Angelman syndrome (AS).
Maternally derived duplications of chromosome 15, specifically 15q11-q13, are associated with an autistic phenotype, whereas paternally derived duplications primarily show normal phenotypes but may manifest neurological features other than autism [
7,
12]. In addition, individuals with PWS with maternal UPD show a higher occurrence of autism compared to those with PWS 15q11-q13 deletions [
13,
14], suggesting that overexpression at maternally expressed genes confers risk for autism [
13,
15]. The E3 ubiquitin ligase gene (
UBE3A) is the only known paternally imprinted gene in the locus, and its maternal allele-specific transcription is limited to postnatal neurons [
16,
17]. While
ATP10A was previously described as maternally expressed [
18,
19], recent studies have shown variable imprinting in humans and lack of imprinting in mouse [
20,
21]. Paternally expressed genes within 15q11-q13 include the splicing factor encoding
SNRPN, necdin (
NDN),
MAGEL2 and several large clusters of small nucleolar RNAs (snoRNAs). A cluster of three receptor subunit genes for the neurotransmitter GABA
A (
GABRB3, GABRA5, GABRG3) are biallelically expressed in control brain tissue samples, but show epigenetic alterations that result in monoallelic expression in a subset of autism cortical samples [
22].
Although increased copy number is generally assumed to increase transcript levels, the epigenetic and neurodevelopmental complexities associated with 15q11-q13 confound this simple explanation of
UBE3A overexpression as the sole molecular cause of the dup15q phenotype. In addition to imprinting, the 15q11-q13 locus is subject to the interchromosomal higher organization of homologous pairing between maternal and paternal alleles [
23‐
25]. Chromosome 15 duplications result in disrupted homologous pairing in dup15q brain tissue samples [
26] and a neuronal cell line model of dup15q [
24]. In addition, disruption of 15q11-q13 pairing by dup15q in neurons has been shown to result in reduced transcript levels of
NDN,
SNRPN,
GABRB3 and
CHRNA7[
24]. In a prior analysis of two dup15q cortical tissue samples, one showed reduced levels of the paternally derived transcripts
SNRPN, snoRNAs,
NDN and the biallelically expressed
GABRB3,
GABRA5 and
GABRG3 transcripts that corresponded to PWS-like behaviors [
26].
To further understand the genetic and epigenetic effects on transcript levels that lead to the pathogenesis of dup15q syndrome, we performed an extensive analysis of a panel of eight dup15q cortical tissue samples as compared to control and idiopathic autism samples. The dup15q samples showed changes in both methylation and transcription levels, with UBE3A showing significantly higher, SNRPN showing significantly lower and GABRB3 showing variable transcript levels as compared to controls. Interestingly, UBE3A transcript positively correlated with maternal allele-specific methylation of the Prader-Willi imprinting control locus (PWS-IC) within the dup15q samples. These results support the hypothesis that elevated UBE3A levels in the brain are a major contributor to the dup15q phenotype but also are consistent with observations that dup15q syndrome results in transcriptional and epigenetic changes that are variable and not based solely on copy number.
Discussion
This paper reports the largest study of dup15q brain samples to date. Our results demonstrate that duplication of the 15q11-q13 region alters the expression not only of
UBE3A, as expected, but also the expression of
SNRPN and
GABRB3 in ways not always predicted by copy number, confirming our prior small-scale study [
26]. Previously,
UBE3A overexpression from the duplicated maternal allele had been hypothesized to be the sole explanation for autism comorbidity in dup15q syndrome as well as the increase in autism spectrum disorder (ASD) phenotypes in PWS maternal UPD compared to deletion cases [
13,
33]. It is important to keep in mind that the PWS-IC is methylated on all maternal alleles, regardless of allele copy number [
34]. Even in studies of various nonneuronal cell lines, however, where
UBE3A is expressed biallelically, increases in
UBE3A transcript in the dup15q cells were observed [
34‐
36]. Our study replicates the prior findings of increased
UBE3A levels in human cortex, showing a twofold increase in dup15q samples. In contrast,
GABRB3 expression was not analyzed in any of the prior studies in cell lines, because
GABRB3 is a neuronally expressed gene.
SNRPN is expressed in nonneuronal cell lines, but researchers in prior studies did not find
SNRPN levels to be different from those of controls in nonneuronal cells [
34‐
36]. In our investigation of dup15q human cortex samples, however,
SNRPN levels were significantly lower than in controls, a result that we did not expect, since all of the samples (control, autism and dup15) should express one copy of the
SNRPN gene from the single paternal allele present. Our results therefore demonstrate the tissue-specific epigenetic complexities associated with dup15q syndrome in humans which simple copy number changes are inadequate to explain.
Epigenetic patterns and mechanisms are often tissue-specific, and the brain shows high levels of DNA methylation despite being primarily nonmitotic in postnatal life [
37]. Our recent genomic analysis of DNA methylation showed large genomic regions that are highly methylated in neurons compared to fibroblasts that span large regions of 15q11-q13 [
38]. Interestingly, in this study, we observed tissue-specific differences in PWS-IC methylation between brain tissues as compared to blood samples analyzed previously [
30] by MS-HRM, with brain tissue showing a higher percentage of baseline maternal allele-specific methylation in controls. The MS-HRM analysis of the PWS-IC upstream of
SNRPN showed that, when normalized to brain, a M:P methylation ratio of 2.9:1 was observed, indicating that the duplications are maternal in origin. The increased methylation observed in dup15q samples is consistent with findings of previous studies in blood from int dup(15) samples showing that the duplication is maternal, not paternal, in origin. However, it is possible that the paternal allele may be methylated at one or more individual bases in the dup15q samples only. The recent discovery of 5-hydroxymethylcytosine (5-hmC) [
39,
40] may be of significance in this regard, because more 5-hmC has been found in brain than in other tissues [
41] and 5-hmC is thought to affect gene regulation through DNA demethylation [
42] or by converting 5-methylcytosine (5-mC) to 5-hmC [
43‐
45]. Further investigation of the methylation status of the PWS-IC in brain samples is needed to determine whether the bisulfite-converted sites are protected by 5-hmC or 5-mC.
UBE3A transcript and protein levels were increased twofold on average in dup15q samples compared to controls in our study, consistent with the hypothesis that there is increased maternal allele-specific expression of
UBE3A in dup15q autism brain. These levels were slightly lower than expected from maternally expressed genes with an average of three maternal alleles, but this may reflect the complex transcriptional and posttranslational regulation of
UBE3A. The function of UBE3A as a transcriptional coactivator has been largely unexplored in the context of human genetic disease, but, in a
Drosophila model of 15q duplication syndrome, elevated levels of an enzymatically defective version of Dube3a were able to induce transcription of the dopamine regulator GTP cyclohydrolase I and elevate dopamine levels in the fly brain [
46]. UBE3A can
trans-ubiquitinate itself
in vivo, leading to self-degradation, supporting the idea that there is an upper limit for UBE3A protein induction that may be reached in as few as two active copies of the duplicated region.
Dup15q sample 6,856 showed a 2.5-fold increase in
UBE3A compared to no significant change as seen previously for a different brain region from this individual [
26]. Brain region differences in transcript levels within the same individual may explain some of the clinical heterogeneity seen within the dup15q syndrome. They may also potentially be explained by the stochastic nature of the epigenetic dysregulation. Interestingly, the epigenetic measure that best correlated with
UBE3A levels in the dup15q brain samples was the level of PWS-IC methylation. Since the correlation was positive rather than negative, we hypothesize that maternal PWS-IC methylation acts as a long-range enhancer of
UBE3A expression. The methyl-binding protein MeCP2 binds to the methylated PWS-IC allele [
25,
31,
47,
48], and
MECP2 mutation has been shown to correspond with reduced UBE3A and GABRB3 levels in human brain [
31]. Therefore, increased binding of MeCP2 to highly methylated PWS-IC in brain may act as a positive transcriptional regulator of
UBE3A and, to a lesser extent,
GABRB3 in human cortex.
In contrast to
UBE3A,
GABRB3 exhibited no significant change in the mean expression in the dup15q cortical samples compared to controls. Instead, significant variability in
GABRB3 levels, as well as an interesting bimodal separation in
GABRB3 levels of the dup15q samples, was observed in dup15q samples. This result is similar to our findings in a prior study of two samples with discordant
GABRB3 levels [
26], as well as the finding of reduced GABRB3 levels in 56% of autism cortex samples [
22].
SNRPN levels were decreased overall in all dup15q samples in this study, which deviates from our previous study result from sample 7014 at a different brain region, BA9, examined previously [
26]. This unexpected result of reduced
SNRPN in dup15q postmortem cortex samples suggests that an increase in maternal dosage of the region epigenetically affects transcription of a paternally expressed gene, possibly in a tissue- or region-specific manner. In contrast to
UBE3A and
GABRB3, which positively correlated with PWS-IC methylation,
SNRPN levels showed a negative correlation with PWS-IC methylation. These results suggest that although maternal methylation of the PWS-IC is repressive to
SNRPN expression, as expected, there appears to be a long-range enhancing effect of PWS-IC methylation on
UBE3A and
GABRB3. Homologous chromosome pairing of maternal and paternal 15q11-q13 alleles occurs in human lymphocytes, neuronal cells and brain [
23‐
25]. Both dup15q brain samples and a neuronal cell culture model of dup15q in SH-SY5Y neuronal cells showed significant disruption of homologous pairing that corresponded to reduced
SNRPN and lower than expected
GABRB3 levels [
24,
26].
Conclusions
This study, together with previous studies of dup15q syndrome, shows that dup15q brain samples are epigenetically complex and that 15q11-q13 transcripts in brain do not behave solely as predicted by copy number. These findings should be important for understanding ASD cases with other
de novo copy number variations on other chromosomes, in particular large duplications [
49]. The bimodal pattern of
GABRB3 deficiencies seen in these 8 dup15q samples may provide some insight into the relationship between dup15q and seizures. Maternal UPD PWS individuals have a higher incidence of seizures than individuals with deletions [
14], suggesting that these people may also have epigenetically induced
GABRB3 deficiency.
GABRB3 and
UBE3A are well-characterized candidate genes for ASD because they are associated with normal brain development and have been shown to be reduced in idiopathic autism, Angelman syndrome and Rett syndrome [
31]. Although these results provide support for the hypothesis that overexpression of the maternally expressed
UBE3A gene in the brain is the primary underlying cause of the ASD phenotype in dup15q, the changes in
GABRB3 and
SNRPN expression not predicted by copy number may also influence the phenotypic variability observed in ASD.
Competing interests
The authors declare that they have no competing interests.
Authors' contributions
HAS carried out the molecular genetic studies and drafted the manuscript. NU carried out the MS-HRM analyses. SWC carried out protein/RNA isolations and Western blot analyses. LTR participated in the study design and coordination and helped to draft the manuscript. JML conceived the study, participated in its design and coordination and helped to draft the manuscript. All authors read and approved the final manuscript.