DNA methylation episignatures: insight into copy number variation

Abstract

In this review we discuss epigenetic disorders that result from aberrations in genes linked to epigenetic regulation. We describe current testing methods for the detection of copy number variants (CNVs) in Mendelian disorders, dosage sensitivity, reciprocal phenotypes and the challenges of test selection and overlapping clinical features in genetic diagnosis. We discuss aberrations of DNA methylation and propose a role for episignatures as a novel clinical testing method in CNV disorders. Finally, we postulate that episignature mapping in CNV disorders may provide novel insights into the molecular mechanisms of disease and unlock key findings of the genome-wide impact on disease gene networks.

Keywords::

• CNV
• DNA methylation
• epigenetic machinery
• EpiSign™
• episignature
• Mendelian disorders

Epigenetic disorders & the epigenetic machinery

Many genes are directly or indirectly linked to epigenetic regulation, particularly those of the epigenetic machinery involved in histone modification, chromatin remodeling and DNA methylation. Epigenetics is the field of genetics involving the study of reversible heritable changes in gene function that occur without changes to the underlying DNA sequence. The epigenetic machinery involves genes encoding for proteins that contribute to the regulation of gene expression at the transcriptional and post-transcriptional level [1]. Components of the epigenetic machinery can be classified as writers, erasers, readers and remodelers. Writers function to modify a region of interest of the genome based on the cell type, as well as on the developmental and metabolic stage of the cell, and they mark individual regions for use or disuse depending on the chromatin state [2]. Writer enzymes include DNA methyltransferases that catalyze the addition of a methyl group to genomic DNA [3]. Eraser genes encode for proteins that remove epigenetic marks and include deacetylase or demethylase enzymes. Reader proteins of the epigenetic machinery recognize the specific epigenetic markers in a locus and may act to recruit other enzymes to this region. The last group are the chromatin remodelers. This class of enzymes uses the energy of ATP to move nucleosomes along the DNA to make regions of chromatin more accessible. The other function of remodelers is to facilitate processes related to protein–DNA interactions [4]. It is important to note that proteins involved in the epigenetic machinery often interact as part of macromolecular structures with complex interdependence and interactivity. The correct functioning of the epigenetic machinery, and thus transcriptional regulation in all processes at the cellular level, particularly during early development, are critical and essential steps in the developmental processes of the body [5]. Therefore, aberrations in epigenetic machinery genes can impair cell function, and development, and ultimately result in epigenetic disorders [2].

In the last decade many new genes associated with the epigenetic machinery have been identified [6]. Bjornsson described the role of a substantial number of these genes in association with 44 Mendelian disorders and their most common overlapping features including neurodevelopmental phenotypes such as intellectual disability (ID), developmental delay and limb/nail abnormalities [7]. These overlapping features create challenges in obtaining a diagnosis with current testing methods [7]. Investigation of the epigenetic machinery and its associated genes is vital to understanding the pathogenesis of these conditions and their phenotypes [7].

Phenotypes reflect the biological information stored in an individual’s genotype and are the sum of all observable characteristics in the genes and the influence of the environment [8]. Over the past decades, this phenotype–genotype association has been demonstrated repeatedly at the molecular level. The relevant biological information can be stored in DNA, RNA or amino acid sequences and the subsequent proteins produced [9]. These proteins regulate the development of an organism from a fertilized egg to an adult individual and ensure proper general cell functioning in the human body. The quality and quantity (dosage) of proteins produced can impact the function of a cell [10]. While sharing the common DNA sequence, different cell types with specific functions and phenotypes contribute to the overall function of a given organism. This is related to the intercellular differences in gene expression, as well as proper temporal and spatial control of this gene expression during development. Therefore, control of gene expression influences cellular differentiation and development and the resulting observed phenotype [10]. Gene expression can be influenced by epigenetic processes including DNA methylation and histone modification. These epigenetic modifications add an important regulatory layer to the dosage requirements for cell-type-specific gene expression, as cells must achieve a balance between the activities of the opposing epigenetic processes, for example, writers and erasers [2].

One type of structural aberration that can impact gene dosage are copy number variants (CNVs), such as deletions or duplications. CNVs involve deletion or duplication of more than 1 kb of genetic material [11] and can be responsible for phenotypic variation and disease susceptibility [12]. CNVs can result in a change in the (total) amount of protein produced (dosage) and, where genes with epigenetic functions are involved, can therefore also systematically impact the balance of epigenetic modifications.

CNVs & dosage sensitivity

Gene dosage sensitivity has been shown to be a major determinant of the pathogenicity of CNVs [13] and previous studies have described a gene balance hypothesis whereby “a stoichiometric balance is maintained among all of the complex gene products in a pathway” [14]. Therefore, dosage effects are the result of the linear relationship between gene copy number and protein product [13], and any deletion or duplication in a gene across the genome that is dosage sensitive could result in phenotypic effects. Multiple ongoing large-scale projects, such as ClinGen Dosage Sensitivity Map (www.clinicalgenome.org) and gnomAD [15,16], aim to accumulate evidence for the dosage sensitivity of the genes in the human genome in order to aid in the clinical assessment of variants. Analogous to this, some genes can be deleted, even in the homozygous state, without apparent phenotypic consequences [13,17]. These genes are described as dosage-insensitive genes. Haploinsufficiency is the result of one copy of a gene being inactivated or deleted and the remaining single gene not producing enough protein to maintain ‘normal’ function [18]. In contrast, triplosensitivity arises when additional copies of a gene increase the protein product and produce a phenotype [19].

Haploinsufficiency of the NSD1 gene (the result of microdeletions or intragenic sequence-level variants) is reported in Sotos syndrome with clinical features including generalized overgrowth, macrocephaly and ID [20]. Alternatively, microduplications of NSD1 have been reported in individuals with a reciprocal Sotos phenotype, presenting with short stature, microcephaly, failure to thrive, seizures and ID [21,22]. It is speculated that gene dosage effects are responsible for the opposing phenotypes observed [21,22]. Similar reciprocal phenotypes have been reported in CNVs at the 1q21.1 [23] and 16p11.2 regions [24–26]. A study by Brunetti-Pierri et al. described dosage effects of the 1q21.1 region associated with head circumference, with deletions resulting in microcephaly and duplications macrocephaly [23]. Deletions of the 16p11.2 region are associated with obesity and macrocephaly [27], whereas duplication of the same region results in low body weight and microcephaly [25]. Dosage of the 7q11.23 region has been shown to affect transcriptional pathways leading to symmetrically opposite expression alterations between deletions and duplications, based on induced pluripotent stem cell modeling [28]. The authors proposed that the symmetrically dysregulated genes within the region may contribute to the opposing disease phenotypes observed in 7q11.23 deletion carriers (Williams–Beuren syndrome) and 7q11.23 duplication carriers (chromosome 7q11.23 duplication syndrome) [28]. Further evidence for the opposing dosage-dependent alterations of these CNVs were reported in a study describing symmetrical dose-dependent DNA-methylation alterations in 7q11.23 deletions and duplications [29]. Deletions of the 7q11.23 region associated with Williams–Beuren syndrome increased DNA methylation, in comparison to the decreased DNA methylation observed in duplication carriers [29].

We postulate that opposing methylation effects may occur in many of the genomic regions where symmetrical deletion and duplications are commonly observed; however, future studies will be required to investigate this further.

Challenges of CNV interpretation

Fragment sizes of deletion and duplication events are quite variable. CNVs can involve a single gene or may consist of multiple genes and/or their regulatory elements. Clinical manifestations may therefore be the result of the effects of disruptions to multiple genes, termed contiguous gene disorders (e.g., 22q11.2 deletion syndrome) or alternatively much of the observed phenotype can be ascribed to the effects of a single causative gene contained within the region (e.g., the NSD1 gene in Sotos syndrome). Overall, these CNVs result in modified or complete absence of protein that may lead to a change in phenotypic presentation [30]. Challenges arise when there is an absence of functional evidence for a CNV or no reports of similar aberrations in the control population databases. Therefore, many CNVs outside of the common genomic disorder regions are classified as variants of uncertain significance (VUS). Further challenges include CNVs that are thought to be pathogenic but have been inherited from apparently unaffected parents [31] and individuals with seemingly identical deletions or duplications that have variable phenotypic features [32]. Reduced penetrance (presence or absence of a phenotype) and variable expressivity (degree and spectrum of observed phenotypes) add an additional layer of difficulty when interpreting and classifying CNVs, as well as in counseling families in the clinic.

Challenges of CNV detection

The detection of CNVs can be carried out using several methods and each method has its own advantages and disadvantages. Clinically relevant CNVs can range from nucleotide-level insertion deletions (indels) to the gain or loss of entire chromosomes [33]. Multiple techniques have been developed and are utilized routinely in clinical laboratories to identify CNVs, as not all CNVs can be detected by the same method. The selection of the method of testing is a combination of national testing guidelines and clinical presentation; however, phenotypic overlap often complicates optimal decision-making regarding the most appropriate first-line choice of a particular diagnostic test [34]. In many instances several of these tests must be performed in a single patient to obtain a diagnosis.

Cytogenetic analysis

Classical cytogenetic methods, including karyotype and fluorescent in situ hybridization (FISH), can detect large CNVs (>3–5 Mb) and microscopic structural features of each chromosome. Karyotyping is the process of pairing and ordering the chromosomes and comparing the banding pattern to identify structural variants. Karyotypes are made with standard staining procedures that show the characteristic structural features (bands) of each chromosome. There are several staining procedures that can be performed to karyotype: G-banding (GTG banding) involves the use of trypsin-Giemsa staining, R-banding (RHG banding) produces the reverse banding pattern from G-banding and uses Giemsa, whereas Q-banding (QFQ banding) uses quinacrine. Additionally, FISH can be performed using fluorescent probes (e.g., spectral karyotyping, multicolour-FISH or targeted FISH) [35,36]. Karyotyping depends heavily on the quality of the sample [37]. If the patient is suspected to have a specific aberration, targeted FISH can be performed with the indicated probes mapping to the region. An advantage of karyotyping is that it is possible to detect apparently balanced translocations and mosaic profiles [38]. It is worth noting, however, that these methods are limited in detection of smaller aberrations (<3 Mb) [35] and are currently primarily used for detection of whole chromosome aneuploidies or balanced rearrangements.

Chromosomal microarray

Chromosomal microarray (CMA) is the first-tier clinical test for individuals with developmental delay, ID and/or congenital anomalies [39]. One type of CMA used to detect chromosomal imbalances is comparative genomic hybridization (CGH). This technique is performed by cohybridizing the fluorescently labeled DNA of a patient with that of a reference control, each labeled with a different color [40]. CGH is used to identify differences in the number of DNA copies of parts of the genome, that is, CNVs [41]. Another type of CMA is the SNP array. This array is more sensitive than CGH, and the DNA probes that are used derive from regions in the genome that show the differences between individuals at a single base pair site thus enabling the detection of polymorphic alleles [42]. The resolution of CGH or SNP arrays differs and depends on probe coverage and location as well as the platform used to perform the array. For example, the Illumina (CA, USA) CytoSNP platform includes more than 850,000 probes enhanced at regions of clinical significance. The resolution of the array-CGH or SNP array is greater than that achieved by the classical cytogenetic technologies and can detect aberrations as small as 10–40 kb [43]; however, CMA cannot detect balanced translocations, inversions, insertions, nucleotide-level changes or low mosaic profiles [37].

Multiplex ligation-dependent probe amplification

Multiplex ligation-dependent probe amplification (MLPA) has been utilized in the detection of CNVs for over two decades. This method allows for quantitative assessment of multiple targeted DNA sequences in a single reaction [44]. Initially, MLPA has been utilized in the assessment of deletions of the 22q11.2 region in DiGeorge/Velocardiofacial syndrome [45]. More recently, many additional kits have become available to detect CNVs in other regions including those associated with Williams–Beuren or Sotos syndromes (MRC Holland, Amsterdam, The Netherlands). MLPA is highly sensitive and can be used to detect CNVs <40 kb, including atypical deletions and duplications which are not routinely detected by classical cytogenetic methods [46]. Similar to FISH, MLPA is also limited to assessment of specific genomic regions. However, MLPA has a rapid turnaround time and is relatively inexpensive compared with other methods [44].

Optical genome mapping

More recently, there has been an emergence of techniques involving optical genome mapping (OGM) that are designed to detect both balanced and unbalanced structural variants in a single platform, including CNVs larger than 500 bp, providing a higher sensitivity than the standard methods (CMA, karyotype, FISH) [47]. In addition, OGM can identify broken, missing, rearranged or extra chromosomes. OGM can also detect structural variants with a low allele fraction [48]. The technique involves labeling ultra-long DNA molecules with a 6 bp general binding motif that creates a pattern that spans the whole genome and is unique to each individual sample. Labeled DNA is then detected through electrophoresis. Every sample has a unique optical genome map, which is aligned to a reference genome, and the differences between those are automatically identified by the software [49]. Limitations of OGM are mostly the result of gaps in the human reference genome and inability to detect breakpoints within large repetitive sequences such as pericentromeric regions and p-arms of the acrocentric chromosomes. The recent mapping of the genome gaps may help to reduce these limitations in the future [50].

Next-generation sequencing & Sanger sequencing

To detect smaller CNVs (1–10 kb) and indels (<1 kb), next-generation sequencing (NGS) can be performed, either as targeted panels or whole exome or genome sequencing (WES and WGS respectively) [51]. Indels can typically be identified by NGS with good sensitivity, but the specificity tends to be low [11]. There is also biased detection of deletions over insertions due to difficulty aligning to the reference sequence. Techniques using read-pair, read-depth, split-read or assembly-based methods have been described to identify indels and CNVs from NGS [52]. Another sequencing method that is widely used is Sanger sequencing. In comparison with NGS, Sanger sequencing only sequences a single DNA fragment at a time. Sanger is mainly used to target smaller genomic regions in a large cohort, for the validation of results from studies performed with NGS or to sequence variable regions [53].

Whole exome sequencing & whole genome sequencing

WES is a DNA sequencing strategy that utilizes NGS techniques to provide a survey of base substitutions and focuses on the exons of 22,000 coding genes, with the output comprising less than 2% of the entire genome. However, a significant proportion of pathogenic variants are located in coding sequences [54,55]. Complexities of interpretation of WES are highlighted by the following: 1) pathogenic variants associated with the phenotype, 2) VUS and 3) secondary findings including pathogenic variants unrelated to the indication for testing.

WGS is a comprehensive method for analyzing entire nonrepetitive genomes including analysis of intronic regions. However, there is limited understanding of the impact of variants, including CNVs, outside of protein coding regions and therefore many of the variants detected in these regions are classified as VUS. Recent studies have shown that CNVs over 50 bp can routinely be detected from targeted NGS gene panels through bioinformatic assessment of the sequencing read depth [56]. Nevertheless, like any clinical test, WES and WGS are not without challenges and limitations and cannot detect large CNVs particularly across repetitive regions. In addition, WES/WGS cannot identify the genomic location of structural variants or low-level mosaicism [51]. Recently, the American College of Medical Genetics and Genomics recommended that WES or WGS be considered as a first- or second-tier test in pediatric patients with congenital anomalies or intellectual disabilities [57].

Long-read sequencing

Long-read sequencing, sometimes referred to as third-generation sequencing, is a method designed to detect DNA sequences up to 140kb, 2–10× coverage depending on the method and platform. Examples of long-range sequencing approaches include ones developed by Pacific Biosciences (PacBio) and Oxford Nanopore Technologies [58]. Nanopore sequencing is a method for long-read cDNA and RNA sequencing and can generally detect up to 10–50 kb. This method involves direct nucleotide assessment without active DNA synthesis. A protein nanopore is stabilized in an electrically resistant polymer membrane and the single stranded DNA passes through this protein nanopore [59]. PacBio is a real-time sequencing method that can detect up to 10–50 kb reads. This method captures sequence data during the replication process and adds a sequence by synthesis strategy that captures a single DNA molecule. PacBio sequencing involves fluorescent light emission that can detect the single nucleotides that are added by the DNA polymerase. Similar to OGM, long-read sequencing utilizes ultra-long high molecular weight DNA, which can be challenging as fresh material or intact cells are required [60]. Long-read sequencing technologies are well suited for genomic, transcriptomic and epigenetic studies [61].

All the methods used in current diagnostic tests involve interrogation of the underlying DNA sequence at the base pair level (e.g., NGS, MLPA or Sanger sequencing) or at the chromosome level (e.g., CMA, FISH or karyotype). Recent studies have reported DNA methylation episignatures as an emerging diagnostic tool that does not directly assess the underlying DNA sequence [62]. Episignatures instead assess the epigenetic consequences of DNA variants [63]. Table 1 provides a summary of the above-described methods.

Table 1. Overview of the different techniques used to detect copy number variants.

Techniques	Strengths	Weaknesses	Detection limit	Ref.
Cytogenetic analysis	Detection of balanced translocations and mosaic profiles	Detection of smaller aberrations	>3.5 Mb	[35–38]
Chromosomal microarray	Resolution is greater than that achieved by classical cytogenetic technologies	Detection of balanced translocations, inversions, insertions, nucleotide level changes or low mosaic profiles	>10–40 kb	[39–43]
Multiplex ligation-dependent probe amplification	Allows for quantitative assessment of multiple targeted DNA sequences in a single reaction	Limited to assessment of specific genomic regions	>40 kb	[44–46]
Optical genome mapping	Detects both balanced and unbalanced structural variants in a single platform and can identify broken, missing, rearranged or extra chromosomes. Optical genome mapping can also detect structural variants with a low allele fraction	Unable to detect breakpoints within large repetitive sequences such as pericentromeric regions and p-arms of the acrocentric chromosomes	>500 bp	[47–50]
Next-generation sequencing and Sanger sequencing	Can be performed either as targeted panels or whole exome or genome sequencing	Can only be used to detect small aberrations	1–10 kb	[51–53]
Whole exome sequencing and whole genome sequencing	Almost the entire genome can be sequenced	Cannot detect large copy number variants particularly across repetitive regions	>50 bp	[54–57]
Long-read sequencing	Long-read sequencing technologies are well suited for genomic, transcriptomic and epigenetic studies	Utilizes ultra-long high molecular weight DNA, which can be challenging as fresh material or intact cells are required	Up to 140kb, 2–10× coverage depending on the method and platform	[58–61]

Episignatures

An episignature is defined as a recurring and reproducible DNA methylation pattern associated with a common genetic or environmental etiology in a specific disorder population [64]. Episignatures are highly sensitive and specific biomarkers that not only detect aberrant DNA methylation but can also provide novel insights into the pathogenesis of genetic conditions as well as highlighting potential targets for future therapies [65]. Episignature testing has shown diagnostic utility in resolving VUS [63,64,66], and detection of an episignature can be used as strong functional evidence for variant reclassification from VUS to likely pathogenic. Alternatively, it should be noted that the absence of a detectable episignature, while suggestive, does not absolutely rule out pathogenicity.

A portion of episignatures that have been identified to date involve genes directly or indirectly associated with the epigenetic machinery that were also described by Bjornsson [7]. Examples include, but are not limited to, the epigenetic writers EZH2 linked to Weaver syndrome [64] and DNMT3A in Tatton–Brown–Rahman syndrome [66], the eraser gene KDM5C in Claes–Jensen syndrome [67], the reader PHF6 in Börjeson–Forssman–Lehmann syndrome [64] and the remodelers SRCAP in Floating-Harbor syndrome [68] and ATRX in alpha-thalassemia mental retardation syndrome [69].

The Illumina microarray (Illumina, CA, USA) is the primary platform used to profile DNA methylation episignatures. This microarray technology provides a genome-wide, cost-effective, and high throughput test that is easily introduced in a laboratory setting with short technical turnaround times. Through bioinformatic applications to the data produced from these microarrays, the DNA methylation profiles can be assessed. EpiSign™ (London Health Sciences Centre Research Inc and Bekim Sadikovic, London, Canada) is a genome-wide DNA methylation diagnostic test that can detect more than 60 rare genetic disorders associated with more than 80 genes [62], with the majority involved in epigenetic regulation. Disorders associated with episignatures that are currently screened with EpiSign V3 test are shown in Figure 1. In addition, EpiSign can also detect several imprinting and repeat expansion disorders including Beckwith–Wiedemann syndrome, Silver-Russell syndrome [70] and fragile-X syndrome [62]. Depending on sample size, effect size, type of trait (continuous, categorical or binary) and distribution of the trait, different methodological approaches can (and should) be applied to study association and predictive power of DNA methylation profiles. For example, as described above, EpiSign is based on a support vector machine algorithm. However, we and other studies optimized their analyses using, among others, elastic net, random forest or other tree-based machine learning approaches [71,72]. In most studies optimization of these models is rather empirical, and the performance of multiple models mostly is evaluated post hoc [73]. The EpiSign workflow to detect novel DNA methylation episignatures is quite unique. The first phase within this workflow focuses on the detection of a disease-associated episignature alone. The second phase of episignature discovery comprises additional training against all other previously detected and validated episignatures, including cycles of optimization if necessary. This approach guarantees the detection of robust episignatures, reflected by both a high selectivity and sensitivity [62].

Figure 1. Current episignatures detectable by EpiSign™ V3.

There are four different categories in this figure: reader, eraser, writer and remodeler proteins. On the outside of the figure there are open circles that denote recessive inheritance and closed circles that denote dominant inheritance. In addition, the disorder phenotypes are shown in different colors, dysmorphic features (green), growth retardation (blue) and intellectual disability (orange).

This figure is adapted with permission from [7], © Bjornsson.

Considering these findings and the previously described DNA methylation profiles seen in patients with deletions and duplications of 7q11.23, 22q11.2 and 22q13.3 [29,65,74], we propose that other CNVs may also exhibit unique diagnostic methylation episignatures.

Episignatures associated with CNVs

DNA methylation arrays, such as the Illumina (CA, USA) Infinium EPIC array utilized in diagnostic episignature testing, have permitted the identification of unique episignatures associated with sequence-level variants in many genes. These discoveries have provided novel insights into the pathogenesis of Mendelian disorders particularly in those genes associated with the epigenetic machinery. However, to date, the utility of assessing DNA methylation profiles that may be the result of larger CNVs has not been widely discussed [75].

Differential methylation in regions associated with some of the most common CNV genomic disorders have been identified. Strong et al. described the role of DNA methylation in the pathogenesis of Williams–Beuren syndrome (7q11.23 deletion) and 7q11.23 duplication syndrome [29]. This CNV region is 1.5 Mb in size and contains approximately 25 genes, of which six have a reported role in epigenetic function. This study concluded that epigenetic mechanisms likely contribute to the complex neurological phenotypes observed in these patients. Another recent study assessed atypical deletions in the 7q11.23 region in order to identify the critical region and genes responsible for the neurodevelopment of patients with Williams–Beuren syndrome [76]. The study concluded that three genes may play an important role: BAZ1B, FZD9 and STX1A [76]. Interestingly, BAZ1B is a tyrosine-protein kinase with an epigenetic regulatory function, as it plays a central role in chromatin remodeling [77]. It is plausible that this gene may be responsible for the aberrant DNA methylation observed in the Strong et al. study; however, further work would be required to assess the DNA methylation patterns in patients with gene-level variants in BAZ1B to confirm this hypothesis.

The interrogation of deletions of different sizes at the 22q13.3 region was also carried out in individuals with Phelan McDermid syndrome [65]. Assessment of DNA methylation profiles in this cohort of individuals identified a critical region responsible for the observed aberrant DNA methylation. This region contained the candidate gene BRD1, a bromodomain-containing protein involved in histone modification. Haploinsufficiency of BRD1 has since been shown to result in abnormal brain development and function due to dysregulation of histone acetylation in mouse models [78]. Investigation of single gene deletions of BRD1 in humans have not been performed but could shed light on the mechanisms responsible for the phenotypic differences observed in Phelan McDermid syndrome individuals.

Differential methylation has also been reported in the 22q11.2 region in DiGeorge/velocardiofacial syndrome [74]. This study also assessed variable sized deletions and identified an episignature in 43 individuals with typical deletions, and two deletions that were more proximal showed the same methylation profile (Figure 2). It is possible that these proximal deletions disrupt enhancer or regulatory regions for genes contained within the typical deletion. Alternatively, deletions may affect the 3D chromatin organization due to disruption of topologically associated domains (TADs) in this region. Differential gene expression has been observed in patients with deletion of 16p11.2 as a consequence of TAD disruption [79]. This suggests that remodeling of chromatin spatial organization plays a key role in the resultant phenotypes. In addition, a DNA methylation episignature was described in a cohort of individuals with autism spectrum disorder and deletions of 16p11.2 [80]. This supports the idea that aberrant DNA methylation patterns may be seen in disorders where TAD disruptions occur.

Figure 2. 22q11.2 deletion syndrome demonstrating that episignatures can be mapped in copy number variant disorders and utilized in clinical testing.

(A) Deletions of the 22q11.2 region: those in red all share a common episignature including the two most proximal deletions. The four deletions represented by the blue horizontal bars do not share the common episignature. (B) Heatmap confirms that 22q11.2 deletion subjects (orange) can be differentiated from age and sex matched controls (blue) using selected CpG probes identified as the episignature. (C) Multidimensional scaling plot confirming the clustering of affected subjects from controls. (D) Support vector machine classifier model. This model was trained using the selected probes for the 22q11.2 episignature, 75% of controls and 75% of other neurodevelopmental disorders samples (blue). The remaining 25% of controls and other disorder samples were used for testing (grey). Plot demonstrates the specificity of the 22q11.2 episignature in differentiating affected cases from other neurodevelopmental disorders with mapped episignatures on the EpiSign™ clinical classifier.

22q11.2DS: 22q11.2 deletion syndrome; ADCADN: Cerebellar ataxia deafness and narcolepsy syndrome; AUTS18: Susceptibility to autism 18; BEFAHRS: Beck-Fahrner syndrome: BFLS: Borjeson–Forssman–Lehmann syndrome; BISS: Blepharophimosis intellectual disability SMARCA2 syndrome; CdLS: Cornelia de Lange syndrome; CHARGE: CHARGE syndrome; Chr16p11.2del: Chromosome 16p11.2 deletion syndrome; CSS: Coffi–Siris syndrome; CSS4: Coffin-Siris syndrome 4; CSS9: Coffin–Siris syndrome 9; Down: Down syndrome; Dup7: 7q11.23 duplication syndrome; DYT28: Dystonia 28; EEOC: Epileptic encephalopathy-childhood onset; FLHS: Floating Harbour syndrome; GTPTS: Genitopatellar syndrome; HMA: Hunter McAlpine craniosynostosis syndrome; HVDAS: Helsmoortel–van der Aa syndrome; ICF: Immunodeficiency-centromeric instability-facial anomalies syndrome; IDDSELD: Intellectual developmental disorder with seizures and language delay; Kabuki: Kabuki syndrome; KDVS: Koolen-De Vries syndrome; Kleefstra: Kleefstra syndrome; LLS: Luscan-Lumish syndrome; MKHK: Menke-Hennekam syndrome; MLASA2: Myopathy lactic acidosis and sideroblastic anemia 2; MRD23: Intellectual developmental disorder 23; MRD51: Intellectual developmental disorder 51; MRX93: Intellectual developmental disorder X-linked 93; MRX97: Intellectual developmental disorder X-linked 97; MRXSA: Intellectual developmental disorder X-linked syndromic Armfield type; MRXSCH: Intellectual developmental disorder X-linked syndromic Christianson type; MRXSCJ: Intellectual developmental disorder X-linked syndromic Claes-Jensen type; MRXSN: Intellectual developmental disorder X-linked syndromic Nascimento type; MRXSSR: Intellectual developmental disorder X-linked syndromic Snyder–Robinson type; PHMDS: Phelan–McDermid syndrome; PRC2: PRC2 complex (Weaver and Cohen-Gibson) syndrome; RENS1: Renpenning syndrome; RMNS: Rahman syndrome; RSTS: Rubinstein–Taybi syndrome; SBBYSS: Ohdo syndrome; Sotos: Sotos syndrome; TBRS: Tatton–Brown–Rahman syndrome; WDSTS: Wiedemann–Steiner syndrome; WHS: Wolf-Hirschhorn syndrome; Williams: Williams syndrome.

This figure is adapted with permission from [74].

The capability of episignatures to differentiate, for example, the 22q11.2 deletion syndrome from other neurodevelopmental disorders, including other CNV disorders such as Williams–Beuren syndrome, is shown in Figure 2D. This sensitivity and specificity of episignatures for the diagnosis of several CNV disorders has been previously described [62]. Taken together, these studies demonstrate the clinical utility of assessing CNVs of differing sizes to uncover novel candidate genes or critical regions that may be responsible for the observed episignature and resultant phenotypes.

Another approach to investigating episignatures in CNVs would be a ‘gene-level up’ method. This method would identify genes with reported roles in epigenetic regulation and assess the DNA methylation profiles of individuals with variants in that specific gene alongside CNVs overlapping the gene of interest. This approach was utilized in mapping a genome-wide DNA methylation episignature in SETD1B-related syndrome [81]. In this study the authors described patients with small gene-level variants and large CNVs in SETD1B, a gene that encodes a SET domain containing protein that is part of a histone methyltransferase complex. This complex primarily targets gene promoters and is highly correlated with gene expression [82]. All patients presented with a similar phenotype that included ID, language delay, epilepsy and behavioral problems such as autism spectrum disorder and anxiety. They also had dysmorphia including full cheeks, full lower lip, macroglossia and tapering fingers. The phenotypes reported are in line with the common features of the Mendelian disorders of the epigenetic machinery described by Fahrner and Bjornsson [2].

Similarly, a genome-wide DNA methylation episignature was reported in individuals with Sotos syndrome [83]. Profiles from individuals with microdeletions of the 5q35 region were assessed alongside gene-level variants in NSD1 for episignature mapping. NSD1 is a histone methyltransferase that can both negatively and positively influence transcription [84]. An episignature was also described for duplications of the same 5q35 region involving NSD1 associated with Hunter–McAlpine syndrome (HMA) [64]. This episignature was able to differentiate HMA from 41 other Mendelian neurodevelopmental disorders. The signature of the 5q35 deletion (Sotos syndrome) included solely hypomethylated loci in contrast to the 5q35 duplication (HMA) which was represented by hypermethylation of the same loci. This mirror effect is most likely the consequence of an opposite dosage of similar genetic loci, that is, the loss of function versus the gain of function [64]. Genes involved in epigenetic regulation are also found in other common CNV regions associated with neurodevelopmental disorders where no episignature has been reported to date. HNRNPU is a gene that codes for a DNA- and RNA-binding protein that links specific DNA elements and plays a role in chromatin organization [85]. It is located within the 1q43q44 deletion syndrome region and has been implicated as the main candidate for the epilepsy component of the disorder’s phenotype [86]. In addition, gene-level variants in HNRNPU detected by WES are described in patients with neurodevelopmental phenotypes [87]. HNRNPU has also been shown to regulate TAD boundaries and TAD interactions and therefore may play a role in chromatin reorganization [85]. Similarly, haploinsufficiency of the histone deacetylase protein HDAC4 has been implicated as the candidate gene in 2q37 deletion syndrome [88]. Individuals with 2q37 deletion syndrome also exhibit phenotypes in line with Mendelian disorders of the epigenetic machinery [2], such as ID and brachydactyly [88].

We hypothesize that disruption of chromatin regulation in regions affected by CNVs, through effects on 3D organization and/or the presence of genes with epigenetic regulatory functions, may result in aberrant DNA methylation profiles. This may be especially true in neurodevelopmental disorders with phenotypes overlapping those previously described by Fahrner and Bjornsson [2]. These DNA methylation episignatures are sensitive and specific biomarkers that could be used for clinical diagnosis [62]. This DNA methylation profile testing approach could alleviate the burden of choosing the appropriate CNV detection method (e.g., WES or CMA) as more CNV disorder episignatures are defined. Taken together, this approach could aid in the refinement of critical regions and genes and further our understanding of the pathogenesis of many genomic disorders that may have implications in the development of targeted therapies in the future.

The potential advantages of episignature mapping in other common CNV disorders was discussed in a recent study by Rooney and Sadikovic [75]. Several of the common CNV regions, associated with neurodevelopmental phenotypes, contain genes with links to the epigenetic machinery (Table 2). This strongly suggests the potential for epigenetic consequences of CNVs in these regions. To date, the majority of these CNV regions do not have mapped episignatures. Of note, chromosome region 1p36 contains the CHD5 gene which is a component of the SWI/SNF chromatin remodeling complex, of which pathogenic variants in several other genes of this complex have been reported to result in aberrant DNA methylation profiles [91]. In contrast, several regions contain genes that, at present, have no reported direct links to the epigenetic machinery. One such region is the 15q11.2 nonimprinted region. It also needs to be acknowledged that the spectrum of episignature disorders is expanding beyond genes directly involved in chromatin and epigenetic regulation. Further work would be required to confirm if changes in the DNA profiles of these 15q11.2 deletion carriers can be detected. Given the evidence of the disruption of 3D chromatin organization via TADs reported in the 16p11.2 deletion region, and effects of TAD disruption on gene expression [79], it’s plausible that deletions of 15q11.2 may still result in an episignature despite an absence of epigenetic machinery genes in this region. Future studies assessing DNA methylation in 15q11.2 deletion carriers would be necessary. An additional challenge in mapping episignatures in CNV disorders may arise in those associated with nonrecurrent breakpoints. Should the episignature be driven by a single gene contained within the region, similar to Sotos syndrome [83], then overlapping genomic coordinates would be pertinent to episignature mapping. Similarly, if changes to the DNA methylation profiles are driven by TADs then differently sized CNVs with differing breakpoints could result in different outcomes of TAD disruption. Again, further work will be required to assess CNV disorders in order to elucidate this further.

Table 2. Epigenetic machinery genes located in common copy number variant regions.

Copy number variant region	Epigenetic machinery genes	Episignature [ref.]
1p36	ICMT, CHD5, TP73, PMRD16, SKI, NOC2L	–
1q21.1	CHD1L	–
1q43q44	HNRNPU, DESI2, ZBTB18, AKT3	–
2q11.2	KANSL3, ARID5A	–
2q13	MIR4435-2HG	–
2q37	HDAC4, D2HGDH, ING5, HDLBP, PASK	–
3q29	PAK2, RNF168	–
4p16.3	NSD2, CTBP1, SLBP, CTBP1, PCGF3	[62]
5p15	ATPSCKMT, MTRR, NSUN2, LPCAT1, BRD9	[89]
5q35	NSD1, UIMC1	[83,64]
7q11.23	METTL27, BUD23, BCL7B, BAZ1B	[29]
8p23.1	TNKS	–
9q34	EHMT1	[90]
10q22.3q23.2	WAPL, DYDC1, MAT1A	–
11p11.2	PHF21A, CD82, ALKBH3	–
11q13.2q13.4	KMT5B	–
15q11.2 (nonimprinting region)	–	–
15q11q13	HERC2	[70]
15q13.3	OTUD7A, KLF13	–
15q24 (BP0–BP1)	–	–
15q24 (BP2–BP3)	SIN3A, COMMD4	–
15q25.2	HDGFL3, BNC1	–
16p13.3	CREBBP	[64]
16p13.11	NDE1	–
16p11.2 distal (BP2–BP3)	SH2B1	–
16p11.2 proximal (BP4–BP5)	PPP4C, HIRIP3, PAGR1, INO80E	[64]
17p13.3	HIC1, SMYD4, MYO1C	–
17p11.2 (Smith–Magenis/Potocki–Lupski region)	ALKBH5, RAI1, PEMT	–
17q11.2	SUZ12	–
17q12	HNF1B, TADA2A, AATF, PIGW	–
17q21.31	KANSL1	[64]
22q11.2 (cat eye region)	CECR2, ADA2	–
22q11.2 (DiGeorge/velocardiofacial region)	THAP7, TRMT2A, COMT, HIRA	[74]
22q11.2 distal type I (D-E/F) region	TOP3B, PPM1F	–
22q13.3	BRD1	[65]
Xp11.22	HUWE1, HSD17B10, SMC1A	–

Utility of episignatures

Assessment of differentially methylated regions (DMRs) identified by DNA methylation arrays can provide information about genes affected by the disrupted DNA methylation elsewhere in the genome that may be contributing to the clinical phenotype observed in patients. Biological pathways downstream of the BAF complex were identified through assessment of DMRs in Coffin–Siris and Nicolaides–Baraitser syndromes associated with variants in ARID1B, SMARCB1, SMARCA4 and SMARCA2 genes [91]. These disorders were shown to have significant overlap in DNA methylation profiles. In addition, gene-set enrichment analyses were performed and showed anatomical and system development as the most significant returned terms [91]. Both syndromes are associated with ID and limb abnormalities and represent a disease spectrum. Of note, these episignatures were also identified in patients with deletions of 6q25 containing the ARID1B gene. This provided novel insights into the pathogenesis of 6q25 deletion syndrome. This is in line with a previous study that assessed a deletion involving ARID1B as the only coding gene in the region and suggested ARID1B as the candidate gene for the disorder [92]. DMRs from a DNA methylation episignature were also assessed in 22q11.2 deletion syndrome and identified hypomethylation of several genes that may contribute to the phenotype [74]. These genes included HOXA2, associated with cleft palate and hearing impairment, and IRF8, known to be involved with immunodeficiency 32B.

HOX genes are also implicated as drivers of embryonic bone, tissue and organ development through gene-set enrichment analysis performed on the DNA methylation episignature in Bohring–Opitz syndrome [93]. The most prominent human phenotype terms identified in the study were related to bone abnormalities of the limbs and limb joints in line with the phenotypes of previously reported epigenetic disorders. Therefore, DNA methylation episignatures may provide novel insights into the mechanisms of genetic disorders and the gene pathways affected. Assessment of genome-wide DNA methylation effects has the potential to unlock a state-of-the-art analysis of gene networks. These insights may provide novel targets for therapy in neurodevelopmental disorders.

Conclusion

Episignature analysis may also have potential utility in investigating reduced penetrance and variable expressivity conditions. To date, episignature mapping has focused on utilizing affected individuals with shared genomic variants to identify a common aberrant DNA methylation profile. It is well documented that several of the common CNV disorders show reduced penetrance and variable expressivity, including 22q11.2 and 16p11.2 deletion syndromes [31,88]. It is postulated that haploinsufficiency alone cannot explain the phenotypic variation observed in these conditions [94]. Epigenetic analysis may help to further elucidate the intrafamilial differences in these families. Further work assessing large cohorts of phenotypically affected and unaffected individuals that carry the same CNV would be required.

Additionally, the detection of an episignature can act as a surrogate for functional studies and help resolve VUS, as has been shown in several studies [63,93,95]. The resolution of VUS minimizes challenges in genetic counseling and provides valuable information for patients and their families.

Future perspective

Bjornsson described 44 genes with specific functions (eraser, writer, remodeler and readers) that are part of the epigenetic machinery and include common overlapping clinical features [7]. This overlap presents challenges to genetic testing using current methodologies, as clinicians must decide between single nucleotide or larger CNV-based techniques. EpiSign™, a recently developed clinical assay utilizing episignatures, has proven utility in differentiating between neurodevelopmental disorders including those associated with CNVs, sequence level variants, repeat expansion disorders and imprinting conditions. Thus, EpiSign™ provides a single platform for the detection of the functional consequences of DNA aberrations, to aid in genetic diagnosis and in resolving VUS. Episignatures have been detected in more genes than initially described by Bjornsson [7] as well as in disorders associated with CNVs, such as DiGeorge/Velocardiofacial syndrome [74]. Therefore, it is necessary to continue to research DNA methylation patterns in all disorder genes and CNVs that could expand the current diagnostic testing platforms and provide novel insights into the pathogenesis of these disorders.

Executive summary

• The epigenetic machinery involves genes encoding proteins that contribute to the regulation of gene expression at the transcriptional and post-transcriptional level.

• In the last decade many new genes associated with the epigenetic machinery have been identified and associated with human disorders.

• Copy number variants (CNVs), such as deletions or duplications, are one type of genetic aberration that has the potential to impact gene dosage.

• Gene dosage sensitivity has been shown to be a major determinant of the pathogenicity of CNVs.

• The detection of CNVs can be carried out using several methods and each method has its own advantages and disadvantages. All the methods used in current diagnostic tests involve interrogation of the underlying DNA sequence at the base pair level (e.g., next-generation sequencing, multiplex ligation-dependent probe amplification or Sanger sequencing) or at the chromosome level (e.g., chromosomal microarray, fluorescent in situ hybridization or karyotype).

• An episignature is defined as a recurring and reproducible DNA methylation pattern associated with a common genetic or environmental etiology in a disorder-specific population. Most episignatures that have been mapped to date involve genes directly or indirectly associated with the epigenetic machinery and overlap with those described by Bjornsson.

• Assessment of differentially methylated regions identified by whole genome DNA methylation analysis can provide invaluable information on additional genes that may be affected elsewhere in the genome that could be contributing to the clinical phenotype observed in patients.

• The detection of an episignature can act as a surrogate for functional studies and help resolve variants of uncertain significance.

• It is necessary to continue to research DNA methylation patterns in all disorder genes and CNVs that could expand the current diagnostic testing platforms and provide novel insights into the pathogenesis of these disorders.

Author contributions

L van der Laan, K Rooney, MMAM Mannens, P Henneman and B Sadikovic contributed to the initial study conception and design. L van der Laan, K Rooney and TMA Trooster designed the figures and tables. The manuscript was written by L van der Laan and K Rooney, and all authors commented on previous versions of the manuscript. All authors read and approved the final manuscript.

Financial & competing interests disclosure

The authors have no relevant affiliations or financial involvement with any organization or entity with a financial interest in or financial conflict with the subject matter or materials discussed in the manuscript. This includes employment, consultancies, honoraria, stock ownership or options, expert testimony, grants or patents received or pending, or royalties.

No writing assistance was utilized in the production of this manuscript.