DNAm is the most common and widely-investigated DNA modification in the human epigenome. While there are several other DNA modifications, the role of DNAm in RDEOs has been elucidated in the greatest detail (Schübeler, 2015). In the human epigenome, DNAm is commonly found at the 5th carbon of cytosine, forming 5-methylcytosine (5 mC), which is most often, but not exclusively, maintained in the context of cytosine-phosphate-guanine (CpG) dinucleotides (Schübeler, 2015). In concert with histone and chromatin modifications, DNAm facilitates maintenance of the state of gene expression within cells (Cedar and Bergman, 2009; Allis and Jenuwein, 2016; Carter and Zhao, 2021). As a simple overview, DNA methyltransferases, such as DNMT3A, DNMT3B, and DNMT3L, establish the pattern of DNAm, which is then maintained by DNMT1 and associated proteins (Schübeler, 2015; Lyko, 2018). The ten-eleven translocation (TET) protein family is responsible for the iterative oxidation of 5 mC–5-hydroxymethylcytosine (5hmC), 5-formylcytosine, and then 5-carboxylcytosine, which can eventually be removed by thymine DNA glycosylase-mediated base excision repair (Wu and Zhang, 2017). Genetic variants that alter the functions of these regulatory proteins can drive global disruption of DNAm level, leading to dysregulation of differentiation and developmental trajectories (Liao et al., 2015; Izzo et al., 2020).

DNA nucleotides are wrapped around an octamer of histone proteins to form the basic unit of chromatin, the nucleosome (Allis and Jenuwein, 2016). Posttranslational modifications, such as acetylation, methylation, phosphorylation, SUMOylation, and ubiquitination, are often found on the N- or C-terminal tails of histones, thereby altering their charge and structure (Stillman, 2018). This can affect how histones interact with each other, DNA, and various other chromatin-binding proteins. For example, the negatively charged acetyl group neutralizes the electrostatic interaction between the lysine-rich histone tail and the negatively charged DNA backbone, which can lead to looser nucleosome packaging and facilitate increased accessibility of DNA to the transcriptional machinery (Shahbazian and Grunstein, 2007). Histone modifications are highly correlated with chromatin states (Ernst and Kellis, 2012), nucleosome spacing and positioning (Valouev et al., 2011), and DNAm patterns (Cedar and Bergman, 2009; Allis and Jenuwein, 2016; Velasco et al., 2021). Together, these epigenetic marks maintain transcriptional control throughout the cell cycle, and are responsive to environmental stimuli (Allis and Jenuwein, 2016).

ATP-dependent chromatin remodeling complexes comprise another major class of epigenetic regulators. There are four main families of chromatin remodelers in eukaryotes: SWI/SNF, ISWI, CHD, and INO80 (Längst and Manelyte, 2015). Complexes in all families share the ability to bind to nucleosomes and break the DNA–histone interaction (Wang et al., 2007). Some remodeler families exhibit more specialized functions, such as the SWI/SNF family complexes that are responsible for nucleosome sliding and increasing DNA accessibility (Cenik and Shilatifard, 2021). Other remodeler families have dynamic functions, such as the INO80 family remodelers, which are involved in histone variant deposition, transcriptional activation, and DNA repair (Längst and Manelyte, 2015). Moreover, as chromatin remodelers alter nucleosome organization and chromatin assembly, their functions also do affect the access of DNMT, TET, and histone-modifying proteins, thus indirectly regulating DNAm and histone modifications (Allis and Jenuwein, 2016). For example, deposition of the histone variant H2A.Z by the chromatin remodeler SRCAP is negatively correlated with DNAm level across plants and animals (Zilberman et al., 2008; Conerly et al., 2010; Zemach et al., 2010). As the chromatin remodelers influence chromatin accessibility, they also play important roles in development, and pathogenic variants of the members of these families often lead to developmental disorders and malignancies (Boerkoel et al., 2002; Wang et al., 2007; Alfert et al., 2019; Cenik and Shilatifard, 2021).

The specificity of epigenetic signatures for each cell type and tissue poses challenges in studying RDEOs not encountered in genomic research, including cellular heterogeneity and difficulties in obtaining the tissue or cell type of interest. As epigenetic mechanisms play important roles in defining cell type and cell function (Carter and Zhao, 2021; Janssen and Lorincz, 2022), RDEOs that are driven by alternative epigenetic states can present atypical cell differentiation and maturation patterns as well as altered epigenetic states in each cell population (Blanco-Betancourt et al., 2004; Lindsley et al., 2016; Kamae et al., 2018; Stremenova Spegarova et al., 2020). These phenomena pose unique problems in analyzing bulk tissue epigenetic data. In heterogeneous tissues, the sources of epigenetic variability can be driven by changes in either cell type composition or the epigenetic regulation within a given cell type, or a combination thereof, and two main methods have been applied to resolve these issues. Many groups have constructed algorithms for computationally estimating cell type proportions in complex tissues. Specifically, novel methods have been developed to deconvolute cell type-specific effects of diseases (Zheng et al., 2018; Rahmani et al., 2019), providing new avenues for delineating the effects of RDEOs on cell type proportion versus cell type-specific effects. It should be noted that these methods have high computing power requirements as they involve complex modeling with interaction terms, and so are more appropriate in cohorts with larger sample sizes or for dimension-reduced data.

In addition to deconvolution, which may suffer from predictor errors or the lack of appropriate reference data sets, many studies have opted to analyze sorted cell samples or to use a single-cell approach. Cell sorting requires the isolation of relevant cell types using techniques such as fluorescence-activated cell sorting (FACS) (Hines et al., 2014; Gasparoni et al., 2018). Alternatively, examination of the patient’s epigenome at the single-cell level can easily distinguish whether the differential regulation is due to compositional changes, cell-specific dysregulation, or a mixture of both (Angermueller et al., 2016; Cheung et al., 2018; Hui et al., 2018; Karemaker and Vermeulen, 2018; Carter and Zhao, 2021). One potential caveat of FACS for purified cells or single-cell analysis is that the tissue dissociation and sorting procedures have been shown to alter transcriptomic profiles (van den Brink et al., 2017). These effects of the sorting process on epigenetic profiles warrant further investigation. Moreover, current single-cell epigenomic technologies result in data degradation and limit the interpretability of the data, so improvements in the experimental and analytic pipelines will be necessary for general application (Ahn et al., 2021; Carter and Zhao, 2021). Overall, single-cell approaches can be used to distinguish the sources of epigenetic variations, which is necessary to gain an understanding of the mechanisms underlying RDEOs.

Animal models, tissue culture, and postmortem tissues are often used to study RDEOs. Some of the most significant strides in RDEO research have been made in model organisms (Mizuguchi et al., 2004; Hagerman et al., 2017; Greenberg et al., 2019; Gamu and Gibson, 2020; Izzo et al., 2020; Huisman et al., 2021; Smith et al., 2021), but not all of this research can be readily translated to humans. For example, mGluR1 signaling blockers showed therapeutic potential for FXS in animal trials, but the effects failed to replicate in clinical trials in adolescents or adults with FXS (Berry-Kravis et al., 2016). These discrepancies may be due to differences in the trajectories of mouse and human brain development and FMRP function (Hagerman et al., 2017). A study examining FMRP binding partners in a human forebrain organoid model showed that, in addition to the presence of many human-specific FMRP binding partners, the organoid model could only be rescued by inhibition of PI3K and not mGluR1 (Kang et al., 2021). The human forebrain model utilized organoids, which are tiny progenitor cell-derived 3D structures in tissue culture that recapitulate basic tissue-level properties (Rossi et al., 2018). Compared to tissue cultures of immortalized cell lines, organoid technology promises to provide higher fidelity representation of complex tissue architecture and cell–cell interactions, with growth and regeneration properties more closely resembling those of primary tissues (Forsberg et al., 2018; Rossi et al., 2018). However, protocols for many organoid types have yet to be optimized; their formation can be unstable and they often fail to reach later stages of development (Rossi et al., 2018). Despite this caveat, organoid culture is an example of a method for mimicking human physiology, and represents an alternative approach for investigating RDEOs in biologically relevant cell models.

Similarly, patient-derived induced pluripotent stem cells (iPSCs) are also valuable for studying RDEOs with cell type specificity. A cocktail of transcription factors is used to convert somatic cells into pluripotent stem cells, which can then be further differentiated to address research questions with tissue specificity (Takahashi et al., 2007; Pozo et al., 2022; Sheridan et al., 2022). As iPSCs retain the donor genotype, they have been widely used to study RDs of genetic origin (Casanova et al., 2014). iPSCs in monolayer culture are more scalable and easier to maintain than 3D organoids, and so serve as better platforms for high-throughput genetic and drug screening (Mellios et al., 2018; Balafkan et al., 2020).

In addition to cell specificity, epigenetic regulation is also sensitive to developmental timing. Most RDEOs have an early onset, as dysregulation of the epigenome affects the trajectory of cell fate decisions and cell identity in a critical period, which is often early embryogenesis (Butler et al., 2012; Lai et al., 2018; Calle-Fabregat et al., 2020; Izzo et al., 2020). Defects in these epigenetic regulatory mechanisms, therefore, lead to specific disease phenotypes that arise within a given developmental window. For example, Dnmt3A and Dnmt3B knockdown are lethal in mice at the embryonic stage, as their functions are critical for epigenetic reprogramming (Okano et al., 1999; Li, 2002). Similarly, null Kmt2d variants in mice are also lethal before embryonic day 10.5, as the protein encoded by this gene is essential for gastrulation (Ashokkumar et al., 2020). There is also evidence that the RDEO disease phenotype can be difficult to reverse past the early developmental period. This phenomenon has been demonstrated in clinical trials of candidate therapeutic agents for FXS, many of which have shown greater efficacy in children than in adolescents or adults (Berry-Kravis et al., 2016; Hagerman et al., 2017). Therefore, experimental conditions that resemble the relevant developmental time point may be crucial to extrapolate the findings for application to patient care.

Animal models are instrumental in understanding the effects of RDEOs on developmental trajectories beyond the molecular phenotypes. In contrast to cell-based models, model organisms can be used to examine morphological and behavioral anomalies as well as system-level dysregulation. In a mouse study, Vallianatos et al. (2020) showed that mutations in key regulators of H3K4 methylation led to changes in dendritic morphology, memory formation, and behavioral aggression. This level of complexity speaks to the unique value of animal studies. Cell-based models, such as iPSCs, can also be used to mimic critical periods (Robinton and Daley, 2012). The process of creating iPSCs drives epigenetic reprogramming, thereby mimicking an embryonic state that can be examined as it is or after differentiation into specialized cell types (Scesa et al., 2021; Pozo et al., 2022). In addition, the use of primary patient samples, such as blood spots or amniotic fluid collected by amniocentesis, can facilitate the identification of RDEOs with biomarkers or be studied later to explore epigenetic dysregulation at an early stage.

Many clinical and experimental studies of RDEOs focus on early developmental time points, potentially due to the early onset of many of these diseases. For example, follow-up of patients into adulthood is comparatively rare, but such studies can be crucial to obtain a detailed clinical picture of RDEOs (Kodra et al., 2018). Studies of the natural history of RDEOs with complex clinical phenotypes can provide insights into the heterogeneity of symptoms and potential comorbidities, and thus inform long-term management and treatment solutions (Hagleitner et al., 2007; Arvio, 2016; Van Remmerden et al., 2020). Similar long-term observations can also be beneficial in animal studies. While animal models are often used to validate the clinical phenotypes observed in patients with RDEOs and validate the genotype–phenotype associations (Cacheiro et al., 2019), they can also be informative regarding potential disease progression. One example outside of RDEO is the report of B-cell malignancy in aging Nfatc2 knockout (KO) mice before the discovery of the first human patient with homozygous LoF NFATC2 variant, who developed B-cell lymphoma as a young adult (May et al., 2014; Sharma et al., 2022). Previous Nfatc2 KO mouse studies mainly focused on skeletal and cartilage defects, as these symptoms manifest early (Ranger et al., 2000; Koga et al., 2005). Comprehensive analysis of RDEO animal models may also be instructive regarding the conditions of patients with corresponding variants. Qualitative studies of patients with RDEO and corresponding animal models can elucidate the effects of epigenetic dysregulation in sensitive tissues throughout the developmental trajectory.

Although there has been a great deal of progress in the discovery and characterization of RDEOs, technical limitations remain major challenges in the field. These challenges are exemplified in the epigenome-wide association study (EWAS), which is the most widely employed method of studying DNAm and uses multiple regression analysis to test for associations between measured DNAm sites and the phenotype of interest (Lappalainen and Greally, 2017; Li et al., 2019). An EWAS typically incorporates hundreds to thousands of samples, with measurement of hundreds of thousands of CpGs (Wahl et al., 2017; Merid et al., 2020). With multiple testing on all sites measured on the commonly employed Illumina EPIC BeadChip array ( ∼ 865 k), around 200 samples, with balanced cases and controls, would be needed to detect a 5% mean difference in methylation status at > 80% power, if a p-value threshold of 0.05 is set after correction for multiple comparisons (Mansell et al., 2019). In RDEOs, the case and control numbers are seldom balanced, leading to a higher type 1 error rate, or the identification of false-positive signals. In addition, the inherent rarity of RDEOs prevents recruitment of sufficient numbers of patients for fully powered analyses, except in the special case of differentially methylated CpGs that show strong and consistent disease-dependent effects, making type II error, or the identification of false-negatives, much more likely.

This problem has led to machine learning strategies and use of polyepigenetic predictors for analyzing RDEO DNAm data. For disease diagnosis, many groups focus on identifying and validating episignatures, i.e., polyepigenetic predictors that use a set of CpGs to estimate disease status (Aref-Eshghi et al., 2019; Sadikovic et al., 2019; Turinsky et al., 2020). To date, useful episignatures have been identified in 65 RDEOs (Levy et al., 2022). Similarly, the EpigenCentral web portal has been created to predict RDEO status using an algorithm that combines the classification results of three machine learning algorithms—penalized logistic regression, random forest, and support vector machine—for seven RDEOs as well as ASD and Down syndrome (Turinsky et al., 2020). However, the prediction accuracy for each disease requires additional validation. These and similar methods have the advantage that clustering-based analysis and machine learning-driven predictors avoid multiple testing, and instead focus on disease classification through the combination of smaller, relevant effects.

Other strategies to mitigate the problem of multiple testing include lessons learned from analysis of ChIP-Seq (chromatin immunoprecipitation followed by sequencing) and ATAC-Seq (assay for transposase-accessible chromatin followed by sequencing) data. Here, the problem of power can sometimes be overcome by summarizing test statistics and developing an analysis pipeline that typically involves chromatin state, motif, and enrichment analysis (Bardet et al., 2012; Steinhauser et al., 2016; Park et al., 2017). A similar strategy has been applied to DNAm data, where genomic regions with similar epigenetic regulation are grouped using R packages, such as CoMeBack and DMRcate (Peters et al., 2015; Gatev et al., 2020). These analyses summarize the high-dimensional data into limited values, so instead of hundreds of thousands of tests, only a few are performed.

There is a strong bias regarding which RDEOs are well-characterized and which are understudied (Ekins, 2017). The less common RDEOs are understudied not only due to funding constraints, but also because of the lack of access to samples. Animal models sometimes do not recapitulate the physiology of RDEOs (Berry-Kravis et al., 2016; Lui et al., 2016; Lui et al., 2018), and primary samples are difficult to obtain because of the rarity of these conditions. For example, both homozygous germline TET2 deficiency and HESJAS have been described in only one report each in the literature (Heyn et al., 2019; Stremenova Spegarova et al., 2020). Alternative bioinformatics approaches can be useful to identify dysregulated epigenetic elements in such cases. With the availability of publicly available data sets, and efforts to create reference epigenomes, such as the Encyclopedia of DNA Elements (ENCODE) project and the NIH Roadmap Epigenomics Project, it is becoming increasingly possible to create a reference epigenome for characterization of RDEOs (Kundaje et al., 2015). While the currently available reference epigenomes are diverse in the types of epigenetic marks examined and the tissues from which the data were generated, they nevertheless have small sample sizes and are low in demographic diversity. The International Human Epigenome Consortium (IHEC) web portal coordinates the creation and publication of reference epigenome maps from seven consortia, including ENCODE, Roadmap, and Blueprint (Bujold et al., 2016). At present, 178 whole-genome bisulfite sequencing (WGBS) data sets in primary human blood cells (including a variety of cell types) and 10 data sets in brain are available on IHEC. By comparing small disease cohorts against large healthy control groups, transcriptomics data have been used to pinpoint disease-associated genes in RDs with variants of unknown significance (Frésard et al., 2019; Ferraro et al., 2020). By expanding on the reference epigenome efforts and applying a similar outlier detection method to identify significantly dysregulated elements, future studies will better identify the downstream pathways that contribute to the pathophysiology of ultra-rare RDEOs.

Further complicating the technical challenges, it can be particularly difficult to elucidate the mechanisms driving RDEOs, as the interaction between genotype and phenotype is often unclear. With the rapid development of whole-genome sequencing technology, it has become increasingly feasible to determine the genetic variants present in patients with aberrant developmental trajectories. However, after sequencing, it can be difficult to determine the variants driving the disease or to delineate the link between the gene of interest and the phenotype in question.

For example, DNMT3A is necessary for de novo DNAm (Lyko, 2018). Differentiated cells derived from Dnmt3a-null mouse hematopoietic stem cells (HSCs) show a global loss of DNAm and deficient repression of HSC-specific genes, demonstrating its role in transcriptional silencing (Challen et al., 2012). These links between DNMT3A LoF and overgrowth phenotype, and between GoF variants and primordial dwarfism can be understood intuitively, but the relations between DNMT3A and brain subfunctions are less well understood. While some studies have demonstrated the necessity of DNMT3A expression for memory formation (LaPlant et al., 2010; Morris et al., 2013; Lavery et al., 2020) and emotional regulation (LaPlant et al., 2010; Elliott et al., 2016; Christian et al., 2020; Lavery et al., 2020), which were further supported by recent findings suggesting a role of non-CpG DNAm dysregulation as a mediator of these effects (Christian et al., 2020; Lavery et al., 2020), the regulatory mechanism has yet to be elucidated. This is mainly because DNMT3A regulates a wide variety of molecular pathways, and changes in its function can sometimes have a cascade effect as DNMT3A targets transcription factors, kinases, or other proteins that in turn exert regulatory functions. Although it is clear that several hundred genes are differentially expressed and methylated in the brains of DNMT3A LoF model mice (Christian et al., 2020; Lavery et al., 2020), the mechanisms underlying the functions of DNMT3A remain to be determined.

To elucidate the mechanisms underlying the regulatory disruptions in patients with RDEOs, it is crucial to identify downstream targets in relevant models and tissues. Targeted epigenomic or transcriptomic profiling in such models can be useful for identifying downstream dysregulated elements. For example, CLIP-Seq (crosslinking immunoprecipitation followed by sequencing) has been used to identify the RNA binding partners of FMRP to study FXS, as FMRP is an mRNA-binding protein. One recent study differentiated four brain cell types consisting of human dorsal and ventral forebrain neural progenitors and neurons, and applied integrated CLIP-Seq and transcriptomic analysis to determine FMRP targets (Li et al., 2020). The results showed that the neurogenesis pathway was consistently disrupted in all four cell types, while cell type-specific differential regulation of genes, such as PIK3CB and SEC24C, was identified and validated in dorsal neurons (Li et al., 2020). Interestingly, upregulation of the catalytic subunit of PI3K, p110b, has been reported to drive deficits in dendritic maturation and cognition in fmr1 knockout mice, and inhibition of p110b reversed the phenotypes associated with FXS (Gross et al., 2010; 2015). These findings were further validated in FXS forebrain organoids, and the results also identified PI3K as the main treatment target (Kang et al., 2021). Taken together, the extensive profiling of FMRP binding partners can generate a list of molecular targets for analysis of their therapeutic potential. This approach can be applied to studying other RDEOs, thus delineating the cascade effects of epigenetic dysregulation and identifying potential therapeutic targets.

The phenotypic and genetic heterogeneity of RDEOs present further challenges in the study of these diseases. In most RDEOs, different patients with genetic variants in the same gene will present a spectrum of clinical phenotypes. Conversely, variants in functionally distinct genes can either lead to the same RDEO or drive similar clinical presentations. While these phenomena complicate the analysis and interpretation of experimental results, a number of tools are available to overcome these challenges. One of the simplest examples illustrating the heterogeneity of RDEO is the apparent reciprocity of the phenotypes associated with GoF versus LoF variants. For example, TBRS and HESJA are driven by LoF and GoF variants of DNMT3A, respectively (Heyn et al., 2019; Smith et al., 2021). The effect of a given variant can be established by a series of functional studies to explore the expression levels of the gene product and downstream target. For example, a global increase in DNAm was demonstrated in germline TET2 LoF disorder, which is consistent with the established function of TET2 in the DNA demethylation pathway (Wu and Zhang, 2017; Stremenova Spegarova et al., 2020).

In addition to the direct contrast of phenotypes attributable to GoF versus LoF variants, the genetic variants can sometimes drive differing phenotypes depending on their location within specific protein domains. For example, variants in two paralogous genes—CREBBP and EP300—are linked to two separate RDEOs; whereas variants outside of exon 30 and 31 of either gene are linked to Rubenstein-Taybi syndrome (RSTS), those within the two exons cause Menke-Hennekam syndrome (MKHK) with distinct clinical phenotypes and DNAm signatures (Bedford et al., 2010; Menke et al., 2016; 2018; Levy et al., 2022). Depending on the functions of the affected domain, variants in the same gene can drive significant phenotypic heterogeneity. Similarly, FLHS is driven by truncation variants in exons 33 and 34 of SRCAP, known as the FLHS locus (Hood et al., 2012; Nikkel et al., 2013; Seifert et al., 2014), whereas non-FLHS SRCAP-related NDD is linked to truncation variants proximal to the FLHS locus (Rots et al., 2021). Although all of the abovementioned SRCAP variants are likely to be non-functional, patients can present with a spectrum of clinical phenotypes, and again functional studies in relevant tissue contexts are warranted to further explore the underlying causes of this phenotypic heterogeneity. The functional differences of the SRCAP variants can drive different epigenome-wide profiles depending on the affected domain. To characterize FLHS SRCAP variants, Greenberg et al. assessed not only the effects of the variants on craniofacial development, but also evaluated the deposition of H2A.Z (a direct downstream target of SRCAP), by ChIP-Seq analyses. FLHS-associated variants were shown to disrupt the nuclear localization of SRCAP and prevent the deposition of H2A.Z.2, a subtype of H2A.Z, demonstrating that the variant is associated with SRCAP LoF (Greenberg et al., 2019). As SRCAP is also a transcriptional activator, RNA-Seq was performed to examine transcriptomic disruption. The application of a similar experimental pipeline to other SRCAP variants may help to resolve the issue of SRCAP-related phenotypic heterogeneity.

Phenotypic heterogeneity can also present as differences in disease onset. While most of the RDEOs discussed above manifest early in life, germline variants of the RDEO-associated genes may also confer a risk of later-onset neurodegenerative diseases. For example, fragile X-associated tremor/ataxia syndrome (FXTAS) is a neurodegenerative disease that affects premutation carriers (55–200 CGG repeats) in the FMR1 gene (Hagerman and Hagerman, 2016; Cabal-Herrera et al., 2020). The symptoms of FXTAS include intention tremor, cerebellar gait ataxia, neuropathic pain, and memory or executive function deficits, which are primarily observed in men older than 50 years (Hagerman and Hagerman, 2016; Cabal-Herrera et al., 2020; Salcedo-Arellano et al., 2020), although heterozygous female premutation carriers may also be affected (Hunter et al., 1998). Interestingly, in contrast to the transcriptional silencing associated with full mutation expansion in FXS, FMR1 mRNA is transcriptionally upregulated by 2–8-fold in patients with FXTAS in comparison to healthy controls (Tassone et al., 2000; 2007). However, this upregulation is accompanied by translational defects, resulting in a decreased FMRP protein level, which is negatively correlated with the number of CGG repeats (Kenneson, 2001; Iliff et al., 2013; Schneider et al., 2020). The sex-specific presentation of the disease indicates that the decreased FMRP dosage contributes to the disease phenotype (Cabal-Herrera et al., 2020). In addition, while the FXTAS variants do not functionally impair early development, the delayed onset also suggests that they cause an accumulation of molecular defects in alternative pathways resulting in neurodegeneration later in life.

Other RDEO-associated genes have also been implicated in neurodegenerative disorders. A recent study showed that rare LoF or non-coding TET2 variants were significantly enriched in populations with early-onset Alzheimer’s disease (EOAD) and frontotemporal dementia (FTD) (Cochran et al., 2020). However, Tet2 loss has also been shown to be neuroprotective in a mouse model of Parkinson’s disease (PD) (Marshall et al., 2020). These conflicting results may be due to differences in experimental setup between studies, or the homeostasis of DNAm regulation may be crucial for brain health. Similarly, DNMT3B and SRCAP variants have also been linked to neurodegenerative disorders, including Alzheimer’s disease (AD), PD, and amyotrophic lateral sclerosis (ALS) (Chesi et al., 2013; de Bem et al., 2016; Chen et al., 2017; Pezzi et al., 2017; Vardarajan et al., 2017). Defects in the DNA damage response have been proposed as potential mechanisms underlying FXTAS (Hagerman and Hagerman, 2016), as DNAm damage and oxidative stress have been shown to induce cellular senescence and neurodegeneration (Coppedè and Migliore, 2015; Martínez-Cué and Rueda, 2020; Gonzalez-Hunt and Sanders, 2021). As FMRPs (Alpatov et al., 2014), TET2 (Feng et al., 2019), DNMT3B (Jin and Robertson, 2013; Shih et al., 2022), and SRCAP (Dong et al., 2014) are all involved in the DNA damage response, the associations between variants in these genes and neurodegenerative disorders may be mediated by altered DNA damaged responses, although further research is needed to establish this potentially shared pathway.

These observations highlight the functional importance of these genes and the harmful effects of associated epigenetic dysregulation. This dysregulation may manifest as epigenetic drift, a phenomenon where the epigenetic profile becomes increasingly variable with age as the epigenetic machinery fails to faithfully maintain regulation through mitosis (Jones et al., 2015; Hernando-Herraez et al., 2019; Bergstedt et al., 2022). It is plausible that variants in key epigenetic regulators exacerbate the rate at which this deterioration occurs. The variants can drive cellular changes that may not translate to clinical symptoms early in life but become apparent at later stages. For individuals with deleterious variants in these genes, biomarkers such as epigenetic clocks—bioinformatics predictors of the biological aging process (Horvath, 2013; Jones et al., 2015; Levine et al., 2018)—may be useful for monitoring disease progression. Acceleration of age-related epigenetic changes was shown to be associated with a diagnosis of ASD using an epigenetic clock trained specifically for children, thereby demonstrating the utility of the clock for detecting altered developmental trajectories in diseases (McEwen et al., 2020). Epigenetic clocks trained in adults can be used to detect aging trajectories later in life (Levine et al., 2018; Lu et al., 2019) and to monitor the processes of epigenetic dysregulation in individuals carrying damaging variants in key epigenetic regulators. This bioinformatics tool should be considered in future RDEO studies across the age spectrum.

Finally, genetic heterogeneity is also commonly observed in RDEOs. Prior to the availability of sequencing technologies to detect disease subtypes with unique genetic origins, diseases such as ICF disorders and PRC2-related overgrowth syndromes were classified as single entities due to the overlap in their clinical phenotypes. In the case of ICF, the functional links between genes driving different subtypes of RDEO are unclear. In ICF, the interactions among drivers of ICF2–4 have been well characterized, while their relations with DNMT3B, the driver of ICF1, remain unknown (Wu et al., 2016; Jenness et al., 2018). ChIP-Seq analyses showed that ICF1 and ICF2-associated variants colocalized to a similar set of genes, which are enriched in pathways involving cellular maintenance, DNA repair, and telomere function (Thompson et al., 2018). The results further suggest that ZBTB24 is necessary for loading of DNMT3B onto DNA (Thompson et al., 2018). While additional studies are required to further characterize these interactions, these observations showed that the identification of shared binding partners and dysregulated effects can be used to probe functional overlap in RDEOs with genetic heterogeneity.

In contrast, genes associated with PRC2-related overgrowth syndromes all encode components of PRC2 (van Mierlo et al., 2019), with defects in each component preventing the normal functioning of the complex. PRC2 subunits have highly coordinated functions: EZH1/2 binds to histone targets and acts as the main catalytic unit (Margueron et al., 2008; Jani et al., 2019); EED provides the epigenetic reader function, propagating transcriptional repression by binding to nucleosomes with H3K27me3 and H3K9me3 and enhancing EZH2 activity (Margueron et al., 2009; Antonysamy et al., 2013; Wu et al., 2013); SUZ12 both stabilizes and recruits PRC2 to chromatin (Choi et al., 2017; Højfeldt et al., 2018); RBBP4/7 are necessary for binding of PRC2 to genomic regions without preexisting H3K27 methylation (Schuettengruber et al., 2017; van Mierlo et al., 2019). Regardless of the distinct roles of each member of core PRC2, they are all essential for its function. Furthermore, PRC2-associated transcriptional regulation involves interactions that extend to other proteins or complexes, including PRC1 (Yu J.-R. et al., 2019), PR-DUB complex (Abdel-Wahab et al., 2012, 2; Balasubramani et al., 2015), and the PRC2 accessory proteins (Oksuz et al., 2018; Glancy et al., 2021). Variants in genes encoding or regulating these elements can also alter the functions of PRC2, and lead to clinical symptoms involving similar pathways to PRC2-related overgrowth syndromes (Gamu and Gibson, 2020). Given that epigenetic regulatory mechanisms are highly interconnected, from DNA and histone modifications to chromatin organization (Cedar and Bergman, 2009; Velasco et al., 2021; Janssen and Lorincz, 2022), it may be relevant to focus on the study of RDEO genes with shared functions. Algorithms can also contribute to the identification of RDEOs with shared clinical and molecular characteristics. GestaltMatcher, for example, uses a deep convolution neural network to classify RDs based on patient‛s facial phenotypes (Hsieh et al., 2022). Data from patients with the same syndromes but different underlying genetic causes were shown to cluster together; for example, subtypes of Kabuki syndrome did not form distinct clusters (Hsieh et al., 2022). Such algorithms can be used to identify RDEOs with overlapping molecular origins based on clinical features. With accumulation of findings regarding related genes and the application of network analyses, the commonality of downstream targets and key regulators should provide valuable insights into the molecular bases of RDEOs.