Introduction

Next-generation sequencing (NGS) techniques like exome/genome sequencing (ES/GS) and RNA-sequencing have boosted the discovery of pathogenic variants in genetic disorders, including neurodevelopmental disorders (NDDs). However, many individuals remain undiagnosed [1]. Variants may be missed due to variant types that current techniques cannot detect, variants in regions of the genome not yet linked to disease, or settings and strategies of genomic data analyses. Moreover, the growing number of genetic variants found with new diagnostic techniques makes it harder to definitively associate them with the clinical presentation, requiring methods for further interpretation.

A type of epigenetic variation undetected by routine diagnostic techniques such as short read ES/GS is DNA methylation (DNAm). DNAm is an epigenetic modification where methyl groups are added to cytosine residues. This modification can regulate gene expression by altering chromosomal structure, DNA conformation, and DNA stability [2]. For instance, hypermethylation of FMR1 occurs due to dynamic CGG repeat expansions and leads to FMR1 silencing, ultimately causing Fragile X syndrome (OMIM#300624) [3]. Another example is hypomethylation of KCNQ1OT1 on chromosome 11p15, which reduces CDKN1C expression and occurs in approximately half of the patients with Beckwith-Wiedemann syndrome (OMIM#130650) [4]. Interestingly, about 5% of Beckwith-Wiedemann syndrome patients exhibit hypermethylation at the maternal IC1, leading to bi-allelic expression of IGF2, normally expressed only from the paternal allele. Thus, the importance of DNAm in disease is well-established in genomic imprinting disorders and metabolic disorders, and is also increasingly recognized in NDDs [2, 5, 6]. The use of DNAm signatures, also known as episignatures, has thereby emerged as a novel diagnostic tool for (neuro)developmental disorders [7, 8].

Distinctive DNAm signatures are described for an increasing number of developmental disorders and can be assessed in DNA isolated from blood. These patterns are often caused by variants in genes encoding proteins involved in epigenetic regulatory mechanisms and chromatin remodeling, resulting in Mendelian disorders of epigenetic machinery (MDEM) [6]. The additional value of DNAm signatures in diagnostic procedures has been increasingly recognized. Contemporary DNAm signatures can be used in a targeted manner to assess whether the presence of an known genetic variant of uncertain clinical significance (VUS) is associated with a disease-linked DNAm signature, which can assist in variant reclassification, or in a comprehensive manner in which DNAm profiles of undiagnosed individuals are compared to that of a panel of disease-linked DNAm signatures, allowing to potentially point to an underlying genetic diagnosis that previously remained unrevealed. For cases with clinical features of MDEM, DNAm signatures were reported to help establish a diagnosis in NGS negative cases and thereby to increase the diagnostic yield [7, 9]. Here, we report an unbiased analysis of the clinical utility of DNAm signatures, focusing on episignatures and imprinting disorders, in a routine clinical setting.

Materials and methods

Study cohort

The cohort consists of 298 consecutive cases from our center who received EpiSignTM testing between February 2019 and June 2023. All included cases were evaluated at the Department of Clinical Genetics, Erasmus MC University Medical Center, Rotterdam, the Netherlands, and examined by a clinical geneticist who provided pre-test counseling for their respective diagnostic workup. All individuals or legal guardians provided written consent to use pseudoanonymized clinical and analysis data. Using genome-wide technologies for diagnostic purposes was previously approved by the Erasmus MC institutional review board (MEC-2012-387).

DNA methylation data were obtained and analyzed via the clinically validated EpiSignTM assay, provided via regular diagnostic services at the Amsterdam UMC [4]. EpiSignTM is a commercial and clinically validated test, which is based on the Illumina EPIC arrays and analyses (1) local hyper- or hypomethylated regions (e.g., for imprinting disorders) and (2) DNAm signatures. During the period analyzed, various versions (V) of the EpiSignTM assay were used. V1 was used from May 2019 to March 2020, V2 from April 2020 to April 2021, V3 from May 2021 to January 2023 and V4 from February 2023 till June 2023. Details about the DNAm signatures and imprinting disorders included in the different versions are available at: (https://genoomdiagnostiek.nl/wp-content/uploads/EpiSIgnv5-Info_2.pdf).

Of the 298 included cases, 75 were requests for targeted analysis for a specific disorder following the identification of a genetic variant in the disorder-associated gene in previous diagnostic genetic testing. Most of these targeted analyses involved VUS (n = 55, 73%), while fewer involved likely pathogenic variants (LP) (n = 11, 15%) or pathogenic variants (n = 9, 12%). In cases where a targeted analysis turned out negative, no systematic complete EpiSign analysis was subsequently performed. 223 of the 298 cases were submitted for comprehensive analysis of all disorders provided by EpiSignTM following prior negative routine diagnostics. These included 2 cases tested with EpiSign v1, 26 cases with EpiSign v2, 140 cases with EpiSign v3, and 55 cases with EpiSign v4, and included both complete analysis of all validated episignatures in the version of EpiSignTM that was used, or early-onset only analysis, which excluded an adult-onset type of autosomal dominant ataxia (ADCADN, OMIM#604121). An overview of all cases is provided in Table S1.

Previous genetic testing across the cohort included specific exome sequencing (ES) panels (multiple congenital anomalies (MCA), intellectual disability (ID), autism, primary immune deficiencies, ciliopathy, mitochondrial DNA, deafness, craniosynostosis) or complete (trio) ES analysis and/or other routine diagnostics including SNP-arrays, Fragile-X syndrome testing, metabolic screening, a variety of diagnostic functional studies and/or clinical RNA sequencing (RNA-seq) (Table S1). Details of local sequencing methods and analysis pipelines are described previously [10, 11].

Uncertain significance (VUS) predictions

Original variant classification was based on the results of the original diagnostics report issued at the time of clinical diagnostics by the accredited diagnostic laboratory of the Clinical Genetics Department of Erasmus MC. For this study, all missense VUS for which a targeted analysis was requested were retrospectively scored for pathogenicity. In silico tools used to assess possible pathogenicity are the CADD-score [12], SIFT [13], PolyPhen [14], Squirls [15], splice-AI [16] and AlphaMissense [17] (Table S1). In addition, we used the ACMG guidelines for variant classification [18, 19] (Table S1), before and after EpiSign testing results. The latter took also into account the current state of knowledge (as of May 2025) in regard to variant allele frequencies and variant entries in ClinVar.

Results

Diagnostic yield: targeted analysis

The yield from the targeted analyses varied depending on the classification of the variant identified. Among the 55 requests to support the pathogenicity and clinical relevance of VUS, 18% (10/55) showed a methylation profile related to the gene of interest (Fig. 1A). Among these 55 VUS, there were 8 truncating, 46 missense, and 1 synonymous variant. Corresponding methylation profiles, found positive for 10 of these variants, were only identified in VUS with truncating effects (2/10) or missense variants with predicted damaging effects by at least 4/5 in silico prediction tools (8/10) (Table S1). Based on these findings, 10 VUS were reclassified as pathogenic variants in accordance with ACMG guidelines (Table 1). This led to diagnoses of Cornelia de Lange syndrome type 1 (NIPBL, OMIM#122470), type 2 (SMC1A, OMIM#300590) and type 4 (RAD21, OMIM#614701), Coffin-Siris syndrome type 1 (ARID1B, OMIM#135900) and type 4 (2x, SMARCA4, OMIM#614609), CHARGE syndrome (CHD7, OMIM#214800), Kabuki syndrome type 2 (KDM6A, OMIM#300867), Claes-Jensen syndrome (KDM5C, OMIM#300534) and intellectual developmental disorder, autosomal dominant 23 (SETD5, OMIM#615761).

Fig. 1: Schematic overview of the study cohort.
figure 1

A Overview of all cases that underwent DNAm signature testing, split for targeted analysis (left) and comprehensive analysis (right). Indicated are the number of cases per group and the various yields. B Pie chart showing the distribution of clinical indications for the 223 cases subjected to comprehensive DNAm signature testing. The number between parentheses indicates the number of cases with a positive DNAm signature per indication group. See Table S2 for a full overview of all indication groups. For simplicity of the pie chart, we merged under the “Others +  isolated phenotypes” (n = 36) category the subgroups “Growth delay” (n = 16), “Epilepsy” (n = 11), “Obesity” (n = 3), “ASS/behavioral anomalies” (n = 2), and “others” (n = 4). Also, we merged the subgroups “DD/ID + short stature” (n = 15) and “DD/ID + tall stature/overgrowth” (n = 3) into “DD/ID + short/long stature” (n = 18). C Pie chart showing the 11 cases with a positive DNAm signature and matching clinical phenotype and their indications for testing. D as (C), but now for the 9 cases with positive DNAm signature, matching clinical phenotype and molecular confirmation.

Full size image
Table 1 Reclassification of variants of uncertain significance using DNAm methylation signatures.
Full size table

Amongst the tested VUS with an initially surprisingly negative DNAm signature (Table S1), there were a de novo 279 kb deletion encompassing SMARCA2 exon 1–25 not showing a BAFopathy DNAm signature; a duplication of the X-linked PHF6 gene detected on an obesity gene panel for which the breakpoints could not be determined and which did not show a Borjeson-Forssman-Lehmann syndrome (OMIM#301900) DNAm signature; a maternally inherited NM_001282531.2(ADNP): c.3213_3216delTGAG, p.(Ser1071Argfs*9) variant not showing a Helsmortel-van der Aa syndrome (OMIM#615873) DNAm signature; a paternally inherited NM_001346813.1 (ARID1B): c.2987-1G>A, p.?, variant not showing a Coffin-Siris type 1 (OMIM#135900) DNAm signature; and a de novo NM_001291415.1 (KDM6A): c.2079+4A>G, p.? variant not showing a Kabuki syndrome type 2 (OMIM#300867) DNAm signature. Potentially, these variants differ from variants used to establish the DNAm signature, explaining their negative results. For example, most pathogenic variants in SMARCA2 are de novo missense variants affecting critical domains of SMARCA2, whereas here a case with a large genic deletion was tested; most pathogenic variants reported for Borjeson-Forssman-Lehmann syndrome are genic mutations in PHF6 whereas here there was initial evidence of a PHF6 gene duplication based on CNV calling of an obesity gene panel performed elsewhere, which later turned out to be instead the presence of an additional X chromosome causing Klinefelter syndrome; and most pathogenic variants reported to cause Helsmortel-van der Aa syndrome are truncating variants affecting the 3’ end of the last exon of ADNP, whereas the variant tested here truncates the gene earlier. Likewise, it is possible that the splice site variant in ARID1B causes an in-frame deletion not removing critical domains of ARID1B, and that the splice site variant in KDM6A does not alter mRNA splicing sufficiently to cause disease-associated DNAm signatures. Indeed, subsequent RNA-seq did show bi-allelic KDM6A expression without evidence of splice alterations. Furthermore, the remaining 40 targeted analyses not showing a positive DNAm signature involved mainly missense variants in the genes ADNP, ARID1A, ARID1B, ATRX, BRWD3 (3x), CHD7 (3x), CHD8 (3x), CREBBP (2x), DNMT3A, DPF2, EHMT1 (2x), EP300, KAT6A, KDM5C, KDM6A (3x), KMT2A (2x), KMT2D (4x), KMT5B, NIPBL (2x), NSD1 (2x), PAF1, SETD1B, SETD5 and TET3 that might not affect critical protein domains explaining the negative DNAm signature.

For likely pathogenic and pathogenic variants, the resulting DNAm signature matched the defined DNAm signatures related to the expected NDD in 91% of the cases (10/11) and 89% of the cases (8/9), respectively (Table S1). DNAm signatures were requested for these (likely) pathogenic variants to further confirm the relation between the identified variant and the clinical features observed, or because the variant was found mosaic. (Likely) pathogenic variants with concordant DNAm signatures were found in the genes ARID1B, CHD8 (4x), H1-4, KANSL1, KAT6B, KDM5C, KMT2A, RAD21, SETD1B, SETD5 (4x), SMC1A, and a duplication of chr1p36.11.

The two discordant (likely) pathogenic variants not showing the expected DNAm signature included a likely pathogenic variant in SETD5 and a pathogenic variant in KAT6B. The NM_001349451.1(SETD5):c.666-3_666-1delinsAA, p.(?), occurred de novo in an individual with developmental delay, ID, and mild dysmorphic features. Variants in SETD5 are associated with Mental retardation, autosomal dominant 23 (OMIM#615761). This could therefore explain most of the phenotype, which led to a clinical diagnosis. The variant causes the usage of an alternative splice site acceptor in exon 10 of SETD5 although this could not be tested by RNA studies. Potentially, the negative DNAm testing result could thus be caused by an alternative disease mechanism, or because the SETD5-related DNAm signature was established based on few SETD5 cases, as noted in the original testing report. The heterozygous pathogenic variant NM_012330.3(KAT6B):c.1354C>T, p.(Arg452*) was identified in an individual presenting with panhypopituitarism, normal psychomotor development, and mild dysmorphic features. The same variant was also identified in the similar affected sibling, which was not tested by EpiSign. Variants in KAT6B are associated with Genitopatellar syndrome (OMIM #606170) and Say-Barber-Biesecker-Young-Simpson syndrome (OMIM#249620), which has overlapping features. Clinical phenotypes include severe psychomotor delay, microcephaly, characteristic facial features, abnormal or missing patellae, urogenital anomalies, and other congenital malformations, but no panhypopituitarism, thus not presenting a phenotypic match. Disease-causing variants in KAT6B are typically found in its final exon 18, but the NM_012330.3(KAT6B):c.1354C>T, p.(Arg452*) variant found here occurs in a region of the gene in which currently no other pathogenic variants have been reported and which is not transcribed in all KAT6B isoforms, potentially explaining the absence of a KAT6B-related DNAm signature.

Diagnostic yield: comprehensive analysis

For the 223 cases submitted for a comprehensive methylation assay following negative routine genome diagnostics, including (panel or full) ES analyses for 96% of cases, a DNAm signature positive for a known disease-specific DNAm or imprinting disorder included in the EpiSignTM assay was observed in 20 cases (9.0%) (Table 2). Of these, the observed profile pointed to a disorder that was (partially) concordant with the phenotypic features and resulted in a clinical diagnosis in 11 out of these 20 cases (55%). DNAm signatures concordant with the described phenotype were observed in 4 cases with developmental delay (DD)/(ID) and the presence of MCA, in two cases with MCA only, in two cases with DD/ID and obesity, in two cases with DD/ID only and in one case with DD/ID with epilepsy (Fig. 1B). For 9 out of 11 of these cases (81.8%) with a concordant phenotype, retrospective analysis identified a causative DNA variant or confirmed independently an imprinting disorder that was unidentified previously, giving rise to a total diagnostic yield of 4.0% (9/223) (Fig. 1A). An overview of the indications for genetic testing, the total number of cases per subcategory, and the number of identified DNAm signatures or imprinting disorders per subcategory is provided in Fig. 1B–D and Table S2.

Table 2 Comprehensive DNAm signature testing can help diagnosing previously unsolved individuals.
Full size table

Molecular diagnoses made included Chung-Jansen syndrome (PHIP, OMIM#617991), Coffin-Siris syndrome type 1 (ARID1A, OMIM#135900), Hunter McAlpine (5q35 duplication, OMIM#601379), CHARGE syndrome (CHD7, OMIM#214800), Wiedemann-Steiner syndrome (KMT2A, OMIM#605130) and Koolen-de-Vries syndrome (KANSL1, OMIM#610443) (Table 2, Table S1). The underlying causative variants identified upon reanalysis were not reported previously due to mosaicism, parentally inherited variants from a parent that only upon reverse phenotyping was identified as being affected, splice site variants outside the regions routinely analyzed in diagnostic procedures, or previous analyses with gene panels that did not include the causative gene at the time of initial analysis. Three additional cases involved imprinting disorders that were not diagnosable with previously performed diagnostic techniques. These included Silver-Russell syndrome (OMIM#180860) based on hypomethylation of the H19 locus and Temple syndrome (OMIM#616222) based on hypomethylation of the 14q32 imprinting region. In the third case, altered methylation of the GNAS locus was found, related to pseudohypoparathyroidism (OMIM#103580). Although this had not manifested clinically until the testing result, low calcium levels were found for which treatment was started. Still, the epilepsy and ID which were the indication for DNAm signature testing remained likely unexplained.

In the remaining two cases (2/11) with a DNAm signature matching the phenotype, no molecular diagnosis has been found to date. These included a case with a DNAm signatures suggesting Cornelia de Lange syndrome and clinical matching phenotype in which ES re-analysis did not identify a mutation in NIPBL, RAD21, SMC3, BRD4, HDAC8 or SMC1A. The second case concerned a DNAm signature indicative of Kleefstra syndrome (OMIM#610253) found in a patient with partially fitting clinical features, but no mutation identified in EHMT1 (or related genes) using ES and RNA-seq.

For 9 cases with positive DNAm signature (9/20, 45%), the profile did not match the initially reported clinical features and did also not provide a clinical diagnosis upon reverse phenotyping (Fig. 1A, Table 2). These included cases with DNAm signatures for Kabuki syndrome (OMIM#147920), UBE2A related disorder (OMIM#300860), Claes-Jensen syndrome (OMIM#300534), KBG syndrome (OMIM#148050) and SETD5 syndrome (OMIM#615761) identified in the same patient, BAFopathy disorders, Wieacker-Wolff syndrome (OMIM#314580), and Rubinstein-Taybi syndrome (OMIM#180849/OMIM#613684), the latter two showing a low confidence DNAm signature. None of these potential diagnoses where clinically matching the phenotype, and extensive ES re-analysis, sometimes followed by clinical RNA-seq did not identify an underlying molecular cause of these DNAm signatures (Table 2, Table S1). In addition, hypomethylation of the imprinting locus KCNQ1OT1 on chr.11p15 as observed in Beckwith Wiedemann syndrome (OMIM#130650), was found in a case without clinical Beckwith Wiedemann syndrome features, presenting with ID and epilepsy. Upon re-analysis, a de novo variant in EEF1A2 was found that provided an alternative diagnosis of developmental and epileptic encephalopathy type 33 (OMIM #616409). Finally, another case showed hypomethylation of the SNURF/SNRPN locus with a paternal chr15q duplication, which was interpreted as not being causative since it was inherited from an unaffected father.

Discussion

We evaluated the clinical utility of DNAm signatures in diagnosing genetic disorders through the analysis of 298 consecutive cases investigated routinely at our clinical genetics department, which included 75 cases with targeted analysis focusing on a specific DNAm signature as pointed to by previous genetic testing results (including 55 VUS) and 223 cases with comprehensive analysis in which previously no potentially disease-explaining variant was identified upon genetic testing. Our findings indicate that DNAm signatures can be a valuable tool in specific contexts, mostly when dealing with variants of uncertain significance (VUS) or cases with high suspicion for a genetic disorder but no diagnosis found in (extensive) previous genetic testing.

Targeted analysis of DNAm signatures is of added value during the evaluation of variants identified with previous genetic testing, especially for providing potentially functional evidence for the effects of VUS. Corresponding DNAm signatures in our cohort were more frequently observed in VUS with a predicted loss of function effect, and in missense variants with predicted damaging effects on protein function. Moreover, the clinical features in these cases typically exhibited at least partial overlap with the features characteristic of the associated disorder, reinforcing the importance to assess both the clinical overlap with known phenotypic features and the pathogenicity of the variants to improve diagnostic accuracy. In total, amongst the 55 VUS that were assessed, targeted DNAm signature testing led to a reclassification of 10 VUS as pathogenic variants (18.2%). Additionally, targeted DNAm signature testing for cases with 11 likely pathogenic and 9 pathogenic variants provided further support for all except two variants. The two discordant results concerned a SETD5 variant that is still considered to likely explain the patient phenotype, and a KAT6B variant that falls outside the common mutation spectrum for the associated disorders, and which is clinically considered unlikely to be causative. Whereas the former variant might not have led to the expected DNAm signature because the SETD5-related signature was established with few cases, shows overlap with the DNAm signature of ANKRD11 and often results in inconclusive results [7], the latter KAT6B variant might have remained negative given the non-typical mutation for the associated disorder. This emphasizes the importance of establishing robust DNAm signatures based on sufficient individuals, which often remains a challenge for rare disorders, and to consider the disease-associated mutation spectrum when interpreting DNAm signature results. Given that a negative DNAm testing result cannot be used to downgrade variants [7], the original classifications of the likely pathogenic SETD5 and pathogenic KAT6B variants were maintained after DNAm testing, although the SETD5 variant is currently more likely considered clinically relevant.

In our comprehensive analysis cohort, DNAm signatures contributed to a confirmed molecular diagnosis in 4.0% (9/223) of previously undiagnosed cases. Two additional clinical diagnoses were made in cases with positive DNAm signatures and matching phenotypes but for which currently no disease-causing DNA variant have been identified, potentially increasing the overall diagnostic yield to 4.9% (11/223). In all these cases, DNAm analysis was performed following extensive earlier genetic testing that had remained negative, including SNP arrays, ES gene panel (37.7%, 84/223) and/or (single/duo/trio) full ES analysis (58.7%, 131/223). The a priori chance of identifying a genetic cause in this cohort was therefore expected to be likely low, although typically in our local clinical routine, patients are selected for additional DNAm signature testing only if a high clinical suspicion of a genetic disorder remains despite earlier negative testing. As the diagnostic yield of ES and SNP-arrays combined is around 30-50% for NDDs, this result shows a small but still relevant complementary contribution of DNAm signatures to genetic diagnostics of previously unsolved cases [1]. As the number of DNAm signatures is rapidly increasing, DNAm signatures are refined when additional affected individuals are tested and generated DNA methylation profiles can be reanalyzed if novel DNAm signatures become available, the diagnostic yield of DNAm signatures will likely become even higher, further increasing the added value of this diagnostic modality [20, 21].

We found a diagnostic yield for targeted analysis of 18.2% for VUS, a diagnostic yield of 4% for comprehensive analysis (only considering cases with matching phenotype and retrospectively confirmed molecular cause) and a discordance rate of 45% upon comprehensive analysis (e.g., where the identified DNAm signature pointed to a disorder which did not match clinically). These numbers differ from previous studies. LaFlamme et al. analyzed 582 cases of unexplained developmental and/or epileptic encephalopathy using the EpiSign v4 classifier and found 7 individuals that harbored positive DNAm signatures concordant with their phenotypes (1.2%, 7/582), of which in 6 cases a pathogenic variant was identified upon reanalysis (1.0%, 6/582) [22]. In a study of 97 well characterized NDD cases divided over various subcohorts, Trajkova et al. found a positive DNAm signature using the EpiSign v3 classifier on targeted testing for 4 out of 18 VUS (22.2%), and upon comprehensive testing in 2 out of 20 unsolved NDD cases with skewed X chromosome inactivation (10%), of which one was molecular confirmed (5%) [23]. In their control cohort, Trajkova et al. found a discordant DNAm signature result for 6 of 34 cases (17.6%). Kerkhof et al. performed 732 targeted and 1667 comprehensive analyses encompassing EpiSign v1 to v3 across several centers offering diagnostic services, finding a diagnostic yield of 32% and 18.3% for both types of analyses, respectively [7]. Their targeted analysis cohort consisted of cases for VUS assessment and/or testing of patients with clinical suspicion of specific DNAm signature disorders. This thus differs from our VUS only cohort and potentially is more reminiscent of our combined cohort of targeted analysis encompassing VUS and (likely) pathogenic variants for which the diagnostic yield is 37.3% (28/75). In their comprehensive analyses, Kerkhof et al. based their diagnostic yield on the number of identified positive DNAm signatures, not considering whether the potential diagnosis matches the clinical phenotypes of the investigated cases. For 56 of their cases, the authors had access to the results of confirmation testing, showing a confirmed molecular diagnosis in 57.1% (32/56, Table 2 of Kerkhof et al.). This contrasts to the 81.2% (9/11 cases) of molecular confirmation that we found in those cases where the DNAm signature associated disorder matched the clinical phenotypes. Although most of these differences in diagnostic yield between studies are thus likely explainable by differences between cohorts, analysis methods and access to detailed patient phenotypes, it is also important to consider differences in the order of genetic testing and when DNA methylation studies are performed during the diagnostic trajectory. Our comprehensive analysis cohort consists mostly of patients that have already undergone extensive prior genetic testing that remained negative, thus using DNAm signature testing as one of the last diagnostic resources. Other centers may use DNAm signatures in earlier stages of the diagnostic process, e.g., before ES, or only for a more limited number of indications. Our series is representative for patients with otherwise unexplained (neuro)developmental disorders from the multi-ethnic populations of Rotterdam routinely seen at a tertiary medical center and includes the comparison between DNAm results and clinical phenotypes.

With the advent of short read WGS, long read WGS, and clinical RNA–sequencing, the landscape of genetic diagnostics is evolving [24, 25]. These advanced sequencing methods will identify a broader range of genetic variations, raising questions about the future role of DNAm signatures. However, cases in our cohort showed that the added value of DNAm signatures is likely to remain significant in the context of inherited VUS and mosaic events, the interpretation of intronic variants, and others that are filtered out or missed by routine sequencing technologies. However, it is known that several DNAm signatures have a higher proportion of inconclusive findings, and are affected by variant type, age and sex [7]. The implementation of diagnostic long-read GS will enable the generation of DNA methylation profiles with DNA sequence information obtained using one technique, leading to a more comprehensive understanding or the genetic variation found and enables the simultaneous refinement of existing DNAm signatures as well as the generation of new DNAm signatures [22, 26, 27].

In conclusion, DNAm signatures are a valuable complement to the diagnostic toolkit for (neuro)developmental disorders, particularly in clarifying the significance of VUS and improving diagnostic accuracy when traditional sequencing methods prove inconclusive. As sequencing technologies advance, integrating genetic and epigenetic data will be crucial for achieving a more precise and comprehensive understanding of developmental disorders.