DNA methylation profiling in Kabuki syndrome: reclassification of germline KMT2D VUS and sensitivity in validating postzygotic mosaicism

Abstract

Autosomal dominant Kabuki syndrome (KS) is a rare multiple congenital anomalies/neurodevelopmental disorder caused by heterozygous inactivating variants or structural rearrangements of the lysine-specific methyltransferase 2D (KMT2D) gene. While it is often recognizable due to a distinctive gestalt, the disorder is clinically variable, and a phenotypic scoring system has been introduced to help clinicians to reach a clinical diagnosis. The phenotype, however, can be less pronounced in some patients, including those carrying postzygotic mutations. The full spectrum of pathogenic variation in KMT2D has not fully been characterized, which may hamper the clinical classification of a portion of these variants. DNA methylation (DNAm) profiling has successfully been used as a tool to classify variants in genes associated with several neurodevelopmental disorders, including KS. In this work, we applied a KS-specific DNAm signature in a cohort of 13 individuals with KMT2D VUS and clinical features suggestive or overlapping with KS. We succeeded in correctly classifying all the tested individuals, confirming diagnosis for three subjects and rejecting the pathogenic role of 10 VUS in the context of KS. In the latter group, exome sequencing allowed to identify the genetic cause underlying the disorder in three subjects. By testing five individuals with postzygotic pathogenic KMT2D variants, we also provide evidence that DNAm profiling has power to recognize pathogenic variants at different levels of mosaicism, identifying 15% as the minimum threshold for which DNAm profiling can be applied as an informative diagnostic tool in KS mosaics.

Introduction

The epigenetic machinery constitutes a network of interconnected components (writers, erasers, readers, and remodelers) that contribute to regulate the spatiotemporal gene expression in a coordinated manner through chromatin remodeling [1]. Among these players, histone modifying enzymes (e.g., methyltransferases, demethylases, acetylases, phosphorylases) regulate chromatin accessibility to transcription factors [1, 2]. Aberrant or defective function of this machinery leads to dysregulated gene expression, causing a group of congenital disorders that are collectively termed as “Mendelian disorders of the epigenetic machinery” [2, 3]. One of these conditions, Kabuki syndrome (KS, MIM: PS147920) is a multiple congenital anomalies/neurodevelopmental disorder (NDD) with an estimated incidence of 1 in 32,000 live births [5]. Originally described by Niikawa et al. [4], KS is characterized by five cardinal features, including postnatal growth restriction, a distinctive facial gestalt (i.e., long palpebral fissures with eversion of the lateral third of the lower eyelid, arched and broad eyebrows with the lateral third displaying notching or sparseness, large and prominent or cupped ears, and short columella with depressed nasal tip), skeletal anomalies (i.e., brachymesophalangy, brachydactyly, fifth digit clinodactyly, and spinal column abnormalities), abnormal finger pads, and developmental delay/intellectual disability (DD/ID), ranging from mild to moderate. The clinical spectrum is overall wide, and presentation tends to evolve over time, often challenging diagnosis [5–7]. Based on these considerations, Makrythanasis and colleagues developed a phenotypic scoring system (MLL2-Kabuki-score [KS-score, hereafter]) to provide a tool to assist the clinical assessment of affected individuals using objective criteria [6]. This systematic scoring system has subsequently been revised and proposed as a clinical tool [8].

KS is caused by inactivating variants in two genes encoding enzymatic components of the epigenetic machinery with complementary writing and erasing functions. KS type 1 (KS1, MIM: 147920) is caused by dominantly acting mutations in the KMT2D/MLL2 (lysine-specific methyltransferase 2D, MIM: 602113) gene [9], which encodes a writer protein with histone H3 lysine 4-specific methyltransferase (H3K4me) activity [10]. KS type 2 (KS2, MIM: 300687) is an X-linked dominant condition caused by pathogenic loss-of-function (LoF) variants in the KDM6A gene (MIM: 300128) [11], which encodes an eraser protein that removes trimethylation from histone H3 lysine 27 (H3K27) [12]. A defective function of either KMT2D or KDM6A results in absent or aberrant balance between methylation of H3K4 and demethylation of H3K27, which is crucial for mono-ubiquitination of histone H2A by recruitment of the PRC1 complex [13]. The identification of a wide spectrum of microdeletions involving these genes or intragenic disruptive deletions/duplications in a subset of KS individuals provides evidence that haploinsufficiency is the molecular basis underlying the disorder [14, 15]. While the nonsense, splice site and frameshift variants underlying KS are spread over the entire coding region of both KMT2D and KDM6A genes, most of the pathogenic missense changes appear to cluster at the C-terminus of the KMT2D protein, which includes five LXXLL motifs required for the interaction of the protein with various nuclear receptors, and the seventh plant homeodomain (PHD), which mediates protein-protein interactions [14–17]. However, the spectrum of variants affecting both genes is overall wide (gnomAD database, https://gnomad.broadinstitute.org/), and their clinical/functional classification still remains a relevant issue. Although the ACMG-AMP guidelines may help in variant classification [18]), the growing incidence of variants of unknown significance (VUS) identified by exome/genome sequencing represents a challenge in the clinical practice (see ClinVar, https://www.ncbi.nlm.nih.gov/clinvar/). In addition, postzygotic mosaicism for pathogenic KMT2D variants is a relevant finding, accounting for up to 2% of individuals with a clinical diagnosis of KS or a suggestive phenotype [19]. Due to the variable extent of cell mosaicism and/or differential tissue distribution, the clinical impact of these variants cannot be established a priori [19–22].

DNA methylation (DNAm) profiling allows to characterize the cumulative methylation patterns occurring at multiple CpG dinucleotides across the genome [23, 24]. It has been demonstrated that pathogenic variants in genes coding for components of the epigenetic machinery result in unique, gene/disease-specific DNAm patterns that are referred to as “episignatures” [25–27]. These DNAm patterns are easily detectable in DNA collected from peripheral blood cells, and have successfully been used to reach diagnosis in patients with unclassified conditions, and as a second-tier tool for the classification of VUS [25, 28–30].

Three disease-specific DNAm signatures are currently available for KS [26, 31, 32]. Here, we compared these episignatures and used the best performing one to classify a variegated set of KMT2D variants, including germline VUS and postzygotic pathogenic mutations. We succeeded in establishing a diagnosis in three subjects carrying a germline VUS in the gene and re-scoring the remaining VUS as not KS-related. We also properly classified 4 out 5 postzygotic pathogenic variants, identifying 15% as the minimum threshold for which DNAm profiling can be applied as an informative diagnostic tool in KS.

Material and methods

Study cohort

A total of 30 individuals with KMT2D variants were enrolled in the study. All patient-derived samples were recruited in the context of routine diagnostic testing. Clinical data, peripheral blood-derived DNA specimens were collected, used, and stored after signed informed consents from the participating subjects/families were secured. All DNA samples and clinical records were pseudonymized. The study was approved by the Ospedale Pediatrico Bambino Gesù (OPBG) Ethical Committee (1702_OPBG_2018). Twelve samples (K1-K12) were heterozygous for a pathogenic (P) or likely pathogenic (LP) KMT2D variant and had a clinical diagnosis of KS; among them, four subjects had previously been reported (K5, K9, K10, and K12) [6, 33]. Six individuals (K1 to K6) were selected as first “validation” set, together with a cohort of 236 age/sex-matched control individuals, including 150 healthy subjects and 86 patients affected with a heterogeneous group of NDDs. These subjects were also used to train the SVM-based classifier to further validate the selected DNAm signature using the remaining subjects with clinical and molecular diagnosis of KS (K7 to K12) and 79 control individuals, including 31 healthy subjects and 48 individuals with a different clinical diagnosis (second “validation” set). Five samples with postzygotic pathogenic KMT2D variants (K13 to K17), and 13 pediatric individuals with clinical features fitting/suggestive of KS and carrying either rare or private KMT2D VUS (K18 to K30), were then tested for classification with respect to the whole control cohort. In the former group, mosaicism was estimated as 10% to 25% in blood-derived leukocytes by assessing the observed allele ratios of parallel sequencing data. Among all tested individuals with germline KMT2D variants, those showing a DNAm pattern fitting the KS-specific signature were then used as final “training” set for the conclusive supervised classification analysis. Mutation analysis approach, and annotation and ACMG classification of the 30 KMT2D variants are provided in Supplementary Table 1; clinical data of each individual are available in Supplementary Table 2. To check the robustness of our findings, we collected and assessed DNAm data from 7 molecularly confirmed KS patients, 1 patient with a KMT2D VUS and 55 healthy controls from Rigshospitalet, Copenhagen (RH). These 7 KS patients were previously found to have DNAm profiles aligning with the DNAm signature of KS [34].

Clinical data were systematically collected for all subjects either diagnosed with KS or having features suggestive for this disorder. To provide an objective clinical assessment, the KS-score [6, 8] was applied. Since a KS-score >4 was previously associated with a definite diagnosis of KS [8, 35], only subjects with a KS-score ≥5 were used in the initial validation set, and only subjects with a KS-score ≥4 were used in the second validation set, except for K9 who carried a de novo truncating variant that had been classified as LP using the ACMG criteria and for whom an incomplete clinical dataset was available.

Methylation analysis

Peripheral blood DNA was extracted using standard techniques. Bisulfite conversion was performed, and samples were analyzed using the Infinium Methylation EPIC BeadChip V.1 (Illumina), according to the manufacturer’s protocol. IDAT files containing methylated and unmethylated signal intensities were imported into R V.4.0.2 for analysis by means of ChAMP V.2.20.1 [36]. Intensity values were corrected for background using the minfi package [37]. Probes located on X/Y chromosomes or known to cross-react with chromosomal locations other than their target regions, containing SNPs at or near the CpG sites, and with a detection p > 0.01 were excluded, resulting in 724,656 high-quality probes that were used in the subsequent analyses. Methylation levels (ß-values) for each sample were compared using the probesets referred to the three currently available DNAm signatures for KS [26, 31, 32], and clustered by means of hierarchical clustering (HC) and multidimensional scaling (MDS), considering the pair-wise Euclidean distances between samples.

The training of the Support Vector Machine (SVM) machine learning (ML)-based classifier was performed with a linear kernel using the e1071 R package V.1.7 and nu-classification option. To determine the best hyperparameter and measure the accuracy of the model, the whole dataset was split into a “training” set (75% of samples) and a “test” set (25% of samples), and a 5-fold cross-validation was performed during the training process. This procedure was repeated four times to verify that each sample was used at least once for testing. Finally, a SMOTE (imbalance R package V.1.0.2.1) oversampling technique was carried out to overcome class imbalance between affected and control individuals used in the training process [38]. Scores from the SVM classifier below 0.25 were considered as control samples, scores between 0.25 and 0.50 were considered inconclusive findings, while scores higher than 0.50 were deemed to behave as KMT2D pathogenic variants.

Results

Validation of the KMT2D DNAm signatures for Kabuki syndrome

Six individuals with bona fide pathogenic/likely pathogenic KMT2D variants and a clinical diagnosis of KS (KS-score ≥5) were initially used as “validation” set (K1-K6) together with a control group (N = 236) to define the best performing KMT2D DNAm signature among the three previously reported (Supplementary Table 1, Supplementary Table 2). Among these, the first two episignatures developed by Butcher et al. and Aref-Eshghi et al. were determined using linear modeling by the identification of differentially methylated probes (DMPs) followed by two different strategies for probeset refinement, mainly based either on the effect size or area under the curve (AUC) and correlation analysis, respectively [26, 31]. The more recently reported DNAm signature by Oexle and colleagues was derived using a different statistical approach and is characterized by a reduced number of DMPs [32]. The three episignatures show only a partial overlap, with the probeset identified by Oexle and colleagues being the most divergent one (Supplementary Fig. 1). As result of MDS and HC analyses, the best performance was reached with the episignature proposed by Aref-Eshghi et al. [26], which was able to discriminate the molecularly confirmed KS individuals from controls with highest sensitivity (Fig. 1). This DNAm signature was further validated by using 6 additional KS individuals with bona fide pathogenic KMT2D variants (K7-K12) and a subgroup of controls (N = 79) by means of both MDS (Fig. 2a, magenta dots) and HC analyses (Fig. 2b). An additional cohort including 7 other bona fide KS cases from an independent external dataset (RH) allowed us to definitively validate this KMT2D episignature (Fig. 2a, b, yellow dots). Using the selected probes, the application of an SVM-based ML classifier, which was trained with samples K1 to K6, allowed to properly score the 13 tested KS cases and all controls with high confidence (Fig. 2c). Thus, the episignature by Aref-Eshghi et al. was applied in the following analyses of DNAm profiles.

Fig. 1. Comparison among different DNAm signatures for Kabuki syndrome.

MDS plots (top) and HC (bottom) obtained from beta-values of different KS-specific CpGs as identified by Butcher and colleagues [31] (a), Aref-Eshghi and colleagues [26] (b), and Oexle and colleagues [32] (c) are reported to compare DNAm profiling-based classification in our internal dataset composed by KS subjects (red) and a control cohort constituted by healthy individuals or subjects affected with other NDDs (light blue).

Fig. 2. Validation of Kabuki syndrome by means of independent datasets.

DNAm signature was tested for robustness using MDS (a) and HC (b) on two different internal cohorts (OPBG, magenta and blue/light blue), and an external dataset (RH, orange and green). Using the selected probes, an SVM-based ML classifier was trained to score all KS cases (red) and controls (light blue) with statistical confidence (c). Controls used as validation set are depicted in blue.

DNAm profiling allows to re-classify subjects with rare/private KMT2D variants of unknown clinical significance

Following the validation steps, K1 to K12 were used as “training” set, to train the SVM-classifier. Then, we proceeded in analyzing the individuals carrying KTM2D variants that had originally been classified as VUS from both internal (13 samples, K18-K30) and external (1 sample, RH-VUS) datasets (Supplementary Table 1, Supplementary Table 2). Among these, three variants (K25, K26 and K30) displayed a KS-specific DNAm pattern (Fig. 3), supporting their classification as KS-causing. Consistently, applying the SVM-classifier, these samples reached an SVM-score of 0.96, 0.85 and 0.90, respectively (Supplementary Table 1). Specifically, K25 refers to a de novo 6-nucleotide duplication (c.15163_15168dupGACCTG), resulting in the insertion of 2 amino acids (p.Asp5055_Leu5056dup), within a C2HC pre-PHD-type 2 zinc finger (residues 5029 to 5069; https://www.uniprot.org/uniprotkb/O14686/entry) located at the N-terminus of a pleckstrin homology domain (PHD), mediating DNA binding (https://prosite.expasy.org/rule/PRU01146). Although preserving the integrity of the reading frame, the variant was predicted damaging in silico (CADD = 21.1). Of note, this duplication and similar indels affecting adjacent residues (i.e., p.His5036_Glu5037delinsGln and p.Glu5038del) have recently been reported as pathogenic in the ClinVar database. On the other hand, K26 is a missense change (c.13622 A > G, p.Lys4541Arg) affecting a disordered region enriched with basic and acidic residues (residues 4503 to 4544), in which pathogenic variants had not previously been reported. A posteriori segregation analyses of the two variants demonstrated their de novo origin, further supporting their clinical relevance. The third case, K30, refers to a missense change (p.Ala5413Val) located within the Set domain, where a subset of amino acid substitutions was proven to cause dramatic consequences on the function of the methyltransferase [39]. As this individual was adopted, segregation analysis could not be performed. The remaining 11 tested samples (K18-24, K27-K29, RH-VUS) clustered within the control group according to the MDS and HC plots. The SVM scoring system provided values < 0.03, thus allowing us to rule out a diagnosis of KS for these subjects (Supplementary Table 1, Fig. 3). The segregation analyses performed a posteriori documented the inheritance of all variants from an apparently healthy parent in all cases with exception of K28 (p.Gly3688Ala), where the variant occurred as a de novo event. The latter did not involve any known functional domain of KMT2D and was predicted to have a low functional impact according to in silico prediction tools (CADD: 12.6; REVEL: 0.306; M-CAP: 0.222). Based on these negative results, exome sequencing was performed in 4 of these individuals, allowing to identify in three of them (K21, K22, and K24) de novo, likely pathogenic variants in different genes (SETD5, MTOR, and KMT2A, respectively) implicated in NDDs that partially overlap with KS (Supplementary Table 1).

Fig. 3. DNAm profiling to test for classification rare KMT2D variants of unknown clinical significance.

MDS (a) and HC (b) analyses on episignature’s CpG site were performed on molecularly unsolved samples from independent datasets: 13 internal VUS (OPBG, green) and 1 external one (RH, black). Using the selected sites, an SVM-based ML classifier was trained to score all KS cases and controls with statistical confidence (c). KS individuals are in red, while controls are depicted in light blue.

DNAm profiling recognizes pathogenic variants at different level of mosaicism

The genome-wide DNAm patterns of variable percentages (10 to 25%, in blood) of mosaic variants in KMT2D were also assessed (Supplementary Table 1). Five individuals (K13-K17) presented with phenotypes suggestive of KS and postzygotic variants, which were all classified as LP/P according to the ACMG criteria. MDS analysis placed them in a variable position between the KS and control clusters, depending on the percentage of mosaicism (Fig. 4, yellow dots). They were followingly re-assessed by an SVM-based ML classifier trained using 15 samples with pathogenic KMT2D variants, including K1-K12 and those that were reclassified from VUS to pathogenic by DNAm profiling (K25, K26, and K30). Mosaicisms from 15% to 25% (K13, K14, K16, and K17) obtained high-confidence pathogenic scores (>0.75) (Supplementary Table 1). On the other hand, the pathogenic variant p.Tyr2060Ter, at a mosaic level of 10%, could not be properly classified by the MDS and HC analyses, and scored <0.1 at the SVM-score. These findings suggest the presence of a threshold of 15% below which the postzygotic event is not recognized by the KS-specific DNAm signature.

Fig. 4. DNAm profiling properly classifies pathogenic variants occurring at different level of mosaicism.

MDS (a), HC (b), and SVM-based ML classifier (c) were used to assess genome-wide DNAm patterns of variable percentages of mosaic pathogenic variants in KMT2D (10 to 25%, depicted in orange). Two different molecularly confirmed KS individuals were also inspected ranging the variant allele frequency down to 20%, 10%, and 5% by dilution with blood-derived DNA obtained from age-matched healthy controls (D1 and D2, depicted in black and gray). KS individuals are in red, while controls are in light blue.

To independently confirm these findings, we modeled mosaic samples (D1 and D2) by using a serial dilution titration starting from blood-derived genomic DNA samples of two different molecularly confirmed KS individuals (K8 and K10), and diluting these samples with blood-derived DNA obtained from age-matched healthy controls to obtain a variant allele proportion of 20%, 10%, and 5%. Parallel sequencing of the relevant KMT2D exons from the titrated DNA samples confirmed the percentages of dilution (data not shown). Based on EPIC DNAm profiling, MDS plot evidenced a gradient in the distribution of synthetic mosaicisms correlating with the extent of dilution of the tested synthetic samples (Fig. 4, black and gray dots). Consistently, the trained ML-classifier was able to properly “functionally” classify the variants at a dilution of 20%, providing an SVM-score similar to that indicated for the K13 and K17 samples (Fig. 4, black and gray dots). DNAm profiling, however, failed in correctly classifying the same variants at higher levels of dilution.

Clinical and molecular re-evaluation of individuals carrying reclassified KMT2D variants and postzygotic mosaicism

We revised the clinical and molecular characteristics of all the individuals carrying a reclassified KMT2D variant by DNAm profiling. The KS-score was adopted to systematically assess the clinical features of these subjects. All the individuals for whom a diagnosis of KS was refused by the SVM-based classifier scored a KS-score ≤3, except for K23 and K27 (KS-score = 4), who were eventually affected with a different disorder, indicating that partial clinical overlap with other disorders might result in an ambiguous KS score (Supplementary Table 2). On the other hand, when considering the cases positive for the KS DNAm signature, 5 out of 8 subjects (K7 to K11) reached KS-score ≤4, possibly due to the clinical variability characterizing KS and/or unavailability of the clinical data, which might result in misclassification by the KS-score algorithm. For example, individual K9 showing the lowest KS-score among the positive samples (KS-score = 3) lacks two of the main facial features of KS (i.e., eversion of the lower lid, and lateral sparse eyebrows). This individual has been molecularly and clinically evaluated at age 5 months. A clinical diagnosis of KS is frequently challenging in infants [33]. It is well acknowledged that the main features of KS are generally recognizable starting from 3 years as the typical facial features may not be present in infants [8]. Individual K7, an 8-year-old child of Italian origin with a de novo pathogenic KMT2D variant (confirmed by DNAm profiling) scored 4 according to the KS-score system. He presented with mild facial features and, again, apparently lacked a major facial feature (i.e., long palpebral fissures). This finding suggests that the DNAm profiling tool may also help to solve any possible misclassification bias of the KS-scoring system, allowing to reach a definite diagnosis of KS. Regarding the three individuals (K25, K26, and K30), who carried VUS that were reclassified as pathogenic by the SVM-based classifier, their KS-scores were ≥6, clinically supporting the present findings.

The clinical characteristics of the 5 individuals with postzygotic pathogenic KMT2D variants were also revised. The 4 subjects with mosaicisms from 15% to 25% (K13, K14, K16, and K17), who were properly classified by the SVM classifier (score ≥0.75), showed a variable phenotype, which was mirrored by a KS-score ranging from 3 to 6. The highest scores were obtained for subjects K13 and K14 (KS-score 6 and 5, respectively), who in turn carried 20% and 25% mosaicism, respectively, for a truncating variant, while a KS-score = 3 was documented for K16 (p.Arg5021Ter, 15% mosaicism). Finally, individual K15, who carried a nonsense variant (p.Tyr2060Ter) at the lowest level of mosaicism (10%) consistently presented with a KS-score of 2, and was not properly classified by our classifier, obtaining an SVM-score of 0.04. Specifically, this individual presented with mild facial features (everted lower eyelids, large dysplastic ears, and a broad nasal root), ID, short stature, but did not show any anomalies in internal organs.

Discussion

In the last decade, genome-wide DNAm profiling has successfully been used to provide supportive evidence for the causative nature of putative disease-causing variants. Despite their growing popularity, poor guidance has been issued on how to generate and use DNAm signatures, with few exceptions in which different strategies have been discussed to evaluate the influence of experimental design, considering training data size, its normalization, and effect size [24, 40]. Episignatures, in fact, are a reflection of the quality of the specific cohorts of cases and controls that used to generate them. Here we tested three currently available DNAm signatures providing evidence of a differential specificity and sensitivity when applied to our tested KS cohort. By using the best performing episignature, we further document the diagnostic utility of this tool in the clinical routine diagnostics, repositioning VUS and conferring a clinical significance to postzygotic mosaicisms, hence helping to solve ambiguous phenotypes of KS. The use of DNAm profiling, in this case, series allowed us to reclassify two missense changes from VUS to pathogenic, confirm the pathogenicity of a small in-frame duplication, and finally exclude the relevance of a group of private/rare variants, including one occurring as a de novo event, in the context of KS.

The large number of KMT2D VUS identified by routine use of massive parallel sequencing techniques (targeted sequencing, WES or WGS) has revealed a critical issue related to proper interpretation of genomic data. In the present series, 13 VUS with conflicting interpretation in ClinVar (K18-K30) and 5 postzygotic mutations (K13-K17) were further analyzed and successfully reclassified by applying this strategy. Out of the 13 KMT2D VUS tested, an in-frame duplication (K25) and a missense change (K26) were reclassified as pathogenic, consistent with their de novo occurrence and the clinical classification of these individuals (KS-score = 7), leading to a definite diagnosis of KS. Similarly, the variant in individual K30 (p.Ala5413Val), for whom parental DNA was not be available, was reclassified as pathogenic (SVM-score = 0.9) allowing us to reach the diagnosis of KS, in line with the clinical phenotype (KS-score = 6).

Our classifier excluded a pathogenic role in KS of 10 VUS, plus another one from an independent external dataset (RH-VUS). Among them, the de novo c.11063 G > C change in K28 cannot warrant causality in the context of KS, resulting in an amino acid substitution predicted to have a consistent low functional impact according to various in silico prediction tools. This individual also had a low clinical score (KS-score = 1). The other reclassified variants (K18-K24, K27 and K29) found in 9 individuals with somehow suggestive phenotypes of KS (KS-score ≤ 4) definitively ruled out a diagnosis of KS. Of note, subsequent segregation analyses demonstrated their inheritance from apparently unaffected parents, confirming benignity. Further genetic testing allowed to establish a definite molecular diagnosis in three of them (K21, K22 and K24). They phenotypically presented with very mild typical KS facial features, and lacked eversion of the lower eyelid, which is the most consistent facial sign of the KS gestalt. They were retrospectively evaluated for the main key features related to their final molecular diagnosis (i.e., obesity and up-slanting palpebral fissures for individual K21, carrying a mutated SETD5 allele; deep set eyes, relative macrocephaly, and wide spaced incisors in K22, carrying a truncating MTOR variant; macrocrania, short stature, delayed bone age, generalized hypertrichosis, including hypertrichosis cubiti in K24, who was heterozygous for a splice site variant in KMT2A). These findings provide evidence of the overlap of some clinical features of KS with those of other disorders, and the need of “functional” confirmation for VUS to exclude a dual molecular diagnosis expressed with a blended phenotype.

Postzygotic de novo mutations resulting in somatic mosaicism are commonly associated with a less severe and/or variable phenotypes compared with those observed in individuals carrying germline pathogenic variants [41], and have been rarely described in KS and associated with atypical KS phenotypes [19, 20]. However, recognizing the pathogenic impact of mosaicism is challenging possibly due to both its distribution in different cells and tissues and its extent. The minimal threshold of mosaicism in blood required to confirm pathogenicity of a variant by DNAm profiling is strictly dependent on the sensitivity and robustness of the generated episignature [24]. Until recently, a dozen of KS mosaic individuals has been reported with variable estimated allele frequencies in blood ranging from 10% to 37% [19–21, 42, 43]. A diagnostic DNAm signature was documented for the less frequently represented allele (close to 11%) and carrying a mosaic nonsense variant [21]. In this context, other researchers attempted an optimization of DNAm signatures specifically aimed at increasing their sensitivity in the case of mosaicisms, but they were currently only able to validate them on in silico models [32]. Here, we tested 5 individuals (K13-K17) with postzygotic KMT2D mutations (from 10% to 25% allelic frequencies in blood) by DNAm profiling. We also tested two sets of serial-diluted samples (5 to 20%) with bona fide pathogenic KMT2D variants to assess the sensitivity of the used DNAm signature-based classifier. Somatic KMT2D variants at levels greater than 20% in blood were properly classified as pathogenic (SVM-score > 0.8). Mosaicism around 15% was found to cause moderate effects (SVM-score close to 0.75). Finally, mosaicism around 10% resulted in SVM-scores <0.1. These findings were confirmed by the analyses performed using the serial-diluted samples. These results consistently indicate that a threshold of 10–15% represents the resolution limit of the assay, at least in the used testing conditions.

The clinical presentation of KS subjects with low level of mosaicism may be milder compared to other individuals carrying the same pathogenic KMT2D variants in the germline. Individuals with a very low percentage of mosaicism may even lack KS clinical features and ID, as previously reported in an apparently normal individual carrying a 7% percentage mosaicism, who gave birth to a daughter affected with KS [44]. In this family, while DNAm profiling properly classified the pathogenic KMT2D variant in the daughter, it failed to detect a KS-specific signature in the mother, in line with the present findings. This consistent finding provides evidence of a technical limit of this technique as a diagnostic test. Indeed, subject K15, who carried a nonsense variant at the lowest level of mosaicism (10%) and classified as negative by DNAm profiling, presented with diluted facial features, short stature and ID that can be retrospectively explained by a KMT2D pathogenic variant. When pondering his reproductive risks, the resolution power of this technique should be considered in the context of genetic counseling.

Our critical revision of the clinically characterized individuals included in this study evidenced that the KS-score might give ambiguous results by either not reaching a suggestive score or by providing false positive results. This could be due to questionnaires scored incorrectly or incompletely for all the requested items (as in the previously published cases K3 and K11 carrying a pathogenic KMT2D variant), or presence of other syndromic conditions that may partially overlap with KS.

In conclusion, the use of DNAm profiling applied for the diagnosis of KS in subjects with inconclusive genetic findings demonstrates that this tool is robust and reliable to rapidly rule out or confirm a clinical suspicion of this disorder, even considering postzygotic occurrence of mutations, taking into account its technical resolution.

Author contributions

MN, AC, and MT conceived the work, interpreted the data, and wrote the manuscript. AC, MF, LP, CC, CN, MH, LC, EM, and ZT contributed to the DNAm analyses. MN, MLD, MG, CC, EP, AB, AN, SM, AS, GM, BD, and MCD collected the clinical and genetic data. MP contributed to the clinical data analyses. All co-authors contributed to the final version of the manuscript.

Funding

This work was supported, in part, by the Italian Ministry of Health (5 × 1000_2019 and RCR-2022-23682289 to MT, and Current Research Funds to AC), and Italian Ministry of Research (FOE_2020 to MT).

Data availability

The genetic and clinical data that support the findings of this work are provided in the manuscript. Previously unreported variants and those that have been reclassified based on DNAm profiling were submitted to ClinVar (SCV004232735 to SCV004232752). The DNAm data are not publicly available due to privacy/ethical restrictions but are available on request from the corresponding author.

Competing interests

The authors declare no competing interests.

Ethical approval

The study was approved by the local Institutional Ethical Committee (ref. 1702_OPBG_2018). Clinical data, and DNA samples were collected and used after signed informed consents from the participating subjects/families were secured.

Supplementary information

The online version contains supplementary material available at 10.1038/s41431-024-01597-9.