Pathogenic variants in KMT2C result in a neurodevelopmental disorder distinct from Kleefstra and Kabuki syndromes

Summary

Trithorax-related H3K4 methyltransferases, KMT2C and KMT2D, are critical epigenetic modifiers. Haploinsufficiency of KMT2C was only recently recognized as a cause of neurodevelopmental disorder (NDD), so the clinical and molecular spectrums of the KMT2C-related NDD (now designated as Kleefstra syndrome 2) are largely unknown. We ascertained 98 individuals with rare KMT2C variants, including 75 with protein-truncating variants (PTVs). Notably, ∼15% of KMT2C PTVs were inherited. Although the most highly expressed KMT2C transcript consists of only the last four exons, pathogenic PTVs were found in almost all the exons of this large gene. KMT2C variant interpretation can be challenging due to segmental duplications and clonal hematopoesis-induced artifacts. Using samples from 27 affected individuals, divided into discovery and validation cohorts, we generated a moderate strength disorder-specific KMT2C DNA methylation (DNAm) signature and demonstrate its utility in classifying non-truncating variants. Based on 81 individuals with pathogenic/likely pathogenic variants, we demonstrate that the KMT2C-related NDD is characterized by developmental delay, intellectual disability, behavioral and psychiatric problems, hypotonia, seizures, short stature, and other comorbidities. The facial module of PhenoScore, applied to photographs of 34 affected individuals, reveals that the KMT2C-related facial gestalt is significantly different from the general NDD population. Finally, using PhenoScore and DNAm signatures, we demonstrate that the KMT2C-related NDD is clinically and epigenetically distinct from Kleefstra and Kabuki syndromes. Overall, we define the clinical features, molecular spectrum, and DNAm signature of the KMT2C-related NDD and demonstrate they are distinct from Kleefstra and Kabuki syndromes highlighting the need to rename this condition.

Pathogenic KMT2C variants cause a disorder, which was named Kleefstra syndrome 2, suggesting its clinical features are highly similar to Kleefstra syndrome. In this study, we delineate the clinical and molecular features of this disorder and show that it is, in fact, different from both Kleefstra and Kabuki syndromes.

Introduction

Mendelian disorders of epigenetic machinery are among the most common forms of neurodevelopmental disorders (NDDs).¹ Humans possess six histone-3 lysine-4 (H3K4) methyltransferases that are divided into three sub-groups, including the trithorax-related subgroup, which consists of KMT2C and KMT2D. These proteins are important components of the epigenetic machinery and are involved in spatiotemporal gene expression regulation.^2,3 KMT2C or KMT2D, together with WDR5, RBBP5, ASH2L, DPY30 (i.e., WRAD subunit), KDM6A, and other proteins, form the Complex Proteins Associated with Set1 (COMPASS) complex² that performs mono- (H3K4me1)⁴ and trimethylation H3K4 (H3K4me3)⁵ at active chromatin sites of gene enhancers and promoters, respectively.

Heterozygous loss-of-function KMT2D (MIM: 602113) variants were identified to cause Kabuki syndrome type 1 (MIM: 147920) in 2010.⁶ However, the consequences of germline KMT2C (MIM: 606833) variants in humans have been identified more recently. In a phenotype-led study, we identified de novo loss-of-function KMT2C variants in individuals with clinical characteristics overlapping Kleefstra syndrome.^3,7 Kleefstra syndrome (MIM: 610253) is caused by haploinsufficiency of euchromatin histone methyltransferase 1 (EHMT1; MIM: 610253) and is characterized by intellectual disability, autism spectrum disorder (ASD), characteristic facial dysmorphisms, and other variable clinical features.⁸ In an independent genotype-led study, we studied variants in histone lysine methyltransferases (KMTs) and demethylases (KDMs) in the Deciphering Developmental Disorders (DDD) cohort and discovered that KMT2C loss-of-function variants result in a neurodevelopmental phenotype with occasional physical anomalies.⁹ However, the clinical and molecular spectrum of the KMT2C-related disorder is largely unknown, as only a few such individuals have been reported. Based on our experience, we hypothesized that the KMT2C-related NDD is a unique entity that is clinically and molecularly different from Kleefstra and Kabuki syndromes.

In this study, we provide comprehensive clinical, molecular, and DNA methylation (DNAm) data for the KMT2C-related NDD based on a large, previously unreported cohort of individuals, as well as demonstrate that this disorder is different from the molecularly related Kleefstra and Kabuki type 1 syndromes.

Material and methods

Cohort recruitment

This study was approved by the institutional review boards of the South Manchester NHS REC, Radboudumc, and the Hospital for Sick Children (research ethics committee approvals 11/H1003/3/AM02, 2011/188, and #1000038847, respectively). Individuals with rare heterozygous pathogenic (P) or likely pathogenic (LP) variants or variants of uncertain significance (VUSs) in KMT2C were identified in clinical diagnostic settings (using standard chromosomal microarray, exome, or genome sequencing¹⁰) or from large NDD research cohorts (the DDD study,¹¹ 100,000 Genomes project,¹² Simons Simplex collection [SSC],¹³ and MSSNG¹⁴). Individuals with rare reported single-nucleotide variants (SNVs), indels, and copy number variants (CNVs) in KMT2C were included. However, CNVs were limited only to those with deletions <1 Mb, which did not affect other predicted haploinsufficient genes. Individuals with additional pathogenic variants in other genes were also included in the study but were excluded from the clinical feature frequency analysis.

All included individuals or their caregivers/legal representatives consented to participate in this research. Genetic and clinical data from individuals were collected via a customized proforma.

Variant analysis

The variants were annotated using the KMT2C The Matched Annotation from the NCBI and EMBL-EBI (MANE) select¹⁵ transcript (GenBank: NM_170606.3, GRCh37). All identified variants were re-classified according to the American College of Medical Genetics (ACMG) guidelines 2015¹⁶ based on the clinical and molecular evidence obtained during this study. The variants and their interpretations were submitted to the ClinVar database (ClinVar: SCV005044911-SCV005044983 and SCV005044991-SCV005045000). Only individuals with LP/P KMT2C variants without known pathogenic variants in other genes were included for further clinical feature analyses.

DNA methylation analysis

Sample processing and DNA methylation (DNAm) signature derivation and analysis were performed similarly as described before.^17,18,19 Briefly, DNA samples underwent bisulfite conversion using the EpiTect Bisulfite Kit (EpiTect PLUS Bisulfite Kit, QIAGEN) following the manufacturer’s protocol. The converted DNAs were then analyzed on the Illumina Infinium Human Methylation EPIC V1 BeadChip (with ∼850,000 CpG sites) at The Center for Applied Genomics (TCAG), Hospital for Sick Children Research Institute, Toronto, Ontario, Canada. The affected individuals’ and controls’ samples were randomly positioned on the arrays.

DNAm analysis was performed using the minfi R package.²⁰ Briefly, the minfi package was used for data preprocessing, quality control, normalization, transformation to β values, and blood cell composition estimation by Housman’s method. Probes with a detection p value >0.05 in >25% of the samples (911 probes), probes located near common polymorphic variants with minor allele frequencies >1% (166,596 probes), non-specific probes (35,613 probes), probes with raw β value equal to 0 or 1 in >25% of samples (240 probes), non-CpG probes (2,377 probes), and chrX and chrY probes (17,485 probes) were removed from the analysis. Additionally, 705 probes behaving like SNVs were removed using the MethylToSNP package.²¹ In total, 645,199 CpG sites remained for the differential methylation analysis. All samples passed the quality control and were suitable for the analysis.

For the signature derivation, the remaining 645,199 CpG sites’ β values were transformed to M values, and differentially methylated CpGs were identified by using a linear regression with monocyte count, batch, and second principal components as covariates using the limma R package.²² Only differentially methylated CpGs with a 10% methylation difference (|Δβ| > 0.10) and false discovery rate (FDR)-adjusted p values <0.05 were selected for the analysis. To remove false positive sites, we further excluded six CpGs with methylation β values in cases and controls that followed a SNP-like pattern. This resulted in a DNAm signature of 51 CpGs with |Δβ| > 0.10. The results were visualized through principal component analysis (PCA) and heatmap plots using the Qlucore Omics software.

Machine learning models and classifications

For the sample classification, we have developed a machine learning model—a support vector machine (SVM) with linear kernel using caret R package²³ as described before.¹⁷ The model was trained on the signature’s CpG sites for the discovery cohorts (16 KMT2C-related NDD and 50 control samples). Because infants have noticeably different blood DNAm profiles, three infant control samples were added to the controls during the model training to increase model’s specificity across different ages. Receiver operating characteristic analysis was used to select the optimal model using the largest value. The SVM model was set to “probability” mode, so the model generated scores ranging between 0 and 1, where scores <0.25 were interpreted as “negative,” >0.5 as “positive,” and 0.25–0.5 as “intermediate.”

The model’s sensitivity and specificity were evaluated using the KMT2C-related NDD and control validation samples, as well as Kabuki type 1 and Kleefstra syndromes samples. Finally, the model was used to classify the 22 testing samples with KMT2C VUSs.

To evaluate the overlap between KMT2C-related NDD DNAm changes with DNAm changes for other disorders of the epigenetic machinery, we analyzed all KMT2C samples on 17 other available DNAm signatures deployed on EpigenCentral portal,²⁴ as described before.²⁵ To evaluate specificity of the KMT2C-related NDD DNAm signature, we have utilized six molecularly confirmed Kabuki syndrome type 1 and six Kleefstra syndrome individuals, as well as 165 healthy controls.

Protein 3D structure analysis

KMT2C is a large (4,911 amino acids) multidomain protein, so there is no solved protein structure of the whole KMT2C protein available. Therefore, for the analysis of possible missense variant effects, we have used (1) solved structures of high-mobility group (HMG) box (PDB: 2YUK), extended plant homeodomain 6 (PHD6) (PDB: 6MLC²⁶), and Su(var)3-9, Enhancer-of-zeste and Trithorax (SET) (PDB: 6KIW,²⁷ PDB: 5F6K, and PDB: 5F59²⁸) domains; (2) homology models for FY-rich (FYR) (based on PDB: 2WZO²⁹) and extended PHD2 (based on PDB: 4NN2,³⁰ PDB: 6U04,³¹ PDB: 7MJU,³² and PDB: 7D8A³³) domains; and (3) ab initio protein models for all missense positions. The homology modeling, analysis, and visualization were performed using YASARA Structure software.³⁴ The ab initio protein models were generated by AlphaFold2³⁵ (multiple modeled overlapping protein fragments of maximal size 1,500 amino acids) and downloaded from the AlphaFold Protein Structure Database.³⁶ The predicted effect was assessed by at least two different protein structures/models, where available (e.g., solved structure and ab initio; two different ab initio structures).

Clinical feature analyses

Consent acquisition, as well as clinical and molecular description of the recruited individuals were provided by their physicians using a standardized proforma (Table S1). However, for statistical analyses, only individuals with LP/P KMT2C variants and without confirmed additional pathogenic variants in other genes were considered. For any feature, we excluded individuals with “UNKNOWN” coding as previously described.³⁷ World Health Organization growth standards per age and sex groups were used to evaluate the growth. Absolute and relative frequencies (expressed as n[%]) were used for describing categorical variables, whereas medians (m) and interquartile ranges (IQRs) were used for describing continuous variables. Chi-square/Fisher’s exact and Mann-Whitney tests were applied to study categorical and continuous variables, respectively, and their associations with sex (females vs. males), variant group (protein-altering variants [PAVs] vs. protein-truncating variants [PTVs]), and inheritance (inherited vs. de novo variants), respectively. Two-tailed, Bonferroni-adjusted p value <0.05 was considered significant for all statistical analyses, which were carried out using the IBM SPSS Version 29 software.

Facial photo analysis

To identify whether KMT2C-related NDD has a facial gestalt, facial 2D photos from 29 out of the 34 individuals with an LP/P KMT2C variant were compared against 29 sex-, age-, and ethnicity-matched individuals with NDDs as controls (sampling different controls five times) by using PhenoScore as described before.³⁸ Not all 34 individuals included in this study could be included in the PhenoScore analyses because a matched control was not found for five individuals.

Briefly, the facial module of PhenoScore utilizes a state-of-the-art convolutional neural network used in facial recognition (QMagFace³⁹) that recognizes facial features and allows for objectively evaluating and statistically comparing different NDD facial gestalts.

Similarly, by using PhenoScore, the photos from these KMT2C-related NDD individuals were compared against the sex-, age-, and ethnicity-matched individuals with either EHMT1 or KMT2D pathogenic variants to objectively evaluate whether the KMT2C-related NDD facial gestalt significantly differs from those with Kleefstra and Kabuki type 1 syndromes, respectively.

Results

KMT2C-related NDD cohort curation and identification of variant spectrum

To systematically analyze the genetic and clinical spectrum of the KMT2C-related NDD, we ascertained 98 individuals with rare reported KMT2C variants, irrespective of their phenotypes through research databases, international collaborations, and previously published individuals.^3,7,9

Seventy-five out of 98 individuals from 67 families had 61 distinct heterozygous KMT2C variants predicted to be PTVs (21 nonsense, 27 frameshift, 5 splice site, and 8 large deletions). These PTVs were classified as P or LP (Figure 1) because loss of function is the currently accepted disease mechanism,^3,7,9 and the gene is intolerant to loss-of-function variants in population (probability of loss of function intolerance [pLI] = 1; observed/expected ratio [o/e] = 0.08 [0.06–0.12]).⁴⁰ Of note, 52 of these P/LP variants were de novo, 11 were inherited (six maternal and five paternal), and in 12 individuals, the inheritance was unknown. Importantly, KMT2C is a large gene consisting of 59 exons (GenBank: NM_170606.3). Ensembl documents 70 transcripts for KMT2C, 31 of which are annotated as protein coding: ranging in amino acid length from 4,968 (ENST00000682283.1) to 83 (ENST00000684278.1). The MANE select transcript (canonical isoform GenBank: NM_170606.3/ENST00000262189.11) encodes a protein of 4,911 amino acids with mean expression across all adult tissues in the Genotype-Tissue Expression (GTEx) database of only 0.83 transcripts per million (TPM), while the most highly expressed transcript (5.08 TPM) is ENST00000485655.2, consisting of only the last four exons of the gene.⁴¹ This, however, might be related to the 3′ bias of the short-read GTEx data.⁴² In our cohort, we found the PTVs are distributed throughout the gene, with the most 5′ and the most distal PTVs being located in the 3^rd and 57^th KMT2C exons, suggesting that in spite of apparently relatively low expression of the canonical (MANE) transcript in adult tissues, PTVs across the gene are likely to be pathogenic. We also did not observe specific clustering of PTVs in the gnomAD database (Figure S1), but based on the skewed variant allele frequency, we noted that some of the observed PTVs are either artifacts due to segmental duplications (e.g., c.2710C>T [p.Arg904Ter] and c.1173C>A [p.Cys391Ter]) or somatic variants (e.g., c.6415C>T [p.Arg2139Ter]), as KMT2C is a known clonal hematopoiesis driver gene.⁴³

Figure 1.

KMT2C variant spectrum among the recruited individuals and population

(A) Likely pathogenic and pathogenic single-nucleotide, indel, and copy number variants identified in this study are shown on the linear structure of KMT2C.

(B) Likely pathogenic and pathogenic as well as benign copy number variants identified in this study in comparison shown on the KMT2C genomic region with segmental duplications (gray), as well as populational deletions and duplications from DGV and gnomAD databases.

We observed a relatively high proportion of individuals with pathogenic CNVs in our cohort (n = 11; 14%; 5/11 are de novo). We, therefore, examined the genomic architecture and found that the KMT2C contains several segmental duplications with high homology (>98%) to sequences elsewhere in the genome (Figure 1B), making this region a “hot spot” for structural rearrangements.⁴⁴ We examined the Database for Genomic Variants (DGV)⁴⁵ and gnomAD⁴⁰ for structural variants affecting this coding part of KMT2C in the control population in the University of California, Santa Cruz (UCSC) genome browser⁴⁶ and identified in total 26 partial or intragenic gains and only a single deletion. This suggests that the partial/intragenic gains are unlikely to be P (if not leading to frameshift). We, however, cannot exclude frequent occurrence of artifacts due to the limitations of current technologies in the regions affected by segmental duplications.

In contrast, non-PTVs have not been classified as P, so 23 out of 98 individuals from 22 families had 22 VUSs (18 missense and two splice variants and two large duplications). Of those, 10 were de novo, six were inherited, and six were of unknown inheritance.

Collectively, these data show that P/LP truncating KMT2C variants are spread across the gene, affecting only MANE or multiple transcripts and that they can be inherited in some cases. Additionally, clinical classification of KMT2C non-truncating variants is currently challenging.

Protein truncating KMT2C variants result in a DNAm signature in peripheral blood

We and others have previously shown that pathogenic variants in genes encoding components of epigenetic regulators are associated with genome-wide DNAm changes⁴⁷ from which disease-associated DNAm signatures can be derived. Therefore, we performed genome-wide methylation screening on peripheral blood-derived DNA from 16 KMT2C-related NDD individuals (discovery cohort) with pathogenic KMT2C PTVs and 50 controls using Illumina methylation EPIC arrays. We identified 51 significant differentially methylated CpG sites at |Δβ| > 0.10 and FDR-corrected p < 0.05 representing the DNAm signature of the KMT2C-related NDD (for simplicity, further called KMT2C DNAm signature) (Table S3). Most of the signature’s CpGs were hypomethylated (42/51, 82%). Interestingly, 4/51 (8%) of the signature’s CpGs sites mapped to two CpG islands of the WT1 gene (MIM: 607102), which were mostly hypomethylated (Figure 2C).

Figure 2.

DNAm signature classification results

(A) PCA plot for the discovery cohort of 16 individuals with KMT2C-related NDD (red) and 50 controls (blue).

(B) Heatmap plot with hierarchical clustering for the discovery cohort of 16 individuals with KMT2C-related NDD (red) and 50 controls (blue), with hypermethylated sites shown in yellow and hypomethylated sites in blue.

(C) Differentially methylated region that maps to several WT1 CpG islands (green) and one of the promoters (purple) (data obtained from the UCSC genome browser) where each dot shows methylation level at a CpG site in the region for each discovery cohort sample, and the line depicts the mean methylation for controls (blue) and cases (red).

(D) SVM model classification results for different groups: 16 discovery KMT2C-related NDD cases (red), 50 controls (magenta), validation KMT2C-related NDD cases (light blue) and their matched controls (dark blue), KMT2C VUSs (black), Kleefstra syndrome individuals (green), and Kabuki syndrome individuals (gray).

(E) KMT2C-related NDD classification results against 17 other DNAm signatures.

Based on the derived KMT2C DNAm signature, we were able to discriminate the discovery KMT2C cases from healthy controls based on PCA and heatmap plots (Figures 2A and 2B, respectively). Next, we trained an SVM model based on the discovery cohort and tested the sensitivity and specificity of the KMT2C DNAm signature using 165 controls without known developmental disorders and 11 additional validation cases with P/LP PTV KMT2C variants. On this SVM model, all controls were classified as negative (SVM values < 0.25, 100% specificity), and 9/11 validation cases were classified as positive (SVM values > 0.5, 82% sensitivity) (Figure 2D). Two validation cases were classified negatively: individual #48 presented with typical clinical features and has a de novo pathogenic frameshift variant c.13107_13108dup (p.Thr4370Argfs^∗11); individual #52 is mildly affected with multiple affected children and has a small LP KMT2C exon 36 and 37 deletion, resulting in frameshift. However, we cannot exclude that these cases are high-level mosaics in blood.

Next, to test whether KMT2C-related NDD-affected individuals share the DNAm signature with other conditions, we analyzed all available samples with an LP/P KMT2C variant on 17 available DNAm signatures deployed at EpigenCentral.²⁴ All KMT2C-related NDD samples were classified negatively on all signatures (Figure 2E).

These results show that significant changes in methylation patterns of DNA derived from the peripheral blood of individuals with P/LP KMT2C variants exist and that these changes can be used to generate a moderate-effect DNAm signature that does not overlap with the other known disorder-specific DNAm signatures.

A DNAm signature can be used to re-interpret KMT2C VUSs

Next, we explored if DNAm signature could enable classification of the KMT2C VUSs by testing DNA samples from 22/23 individuals (20 unrelated and one proband-father pair) with 21 distinct KMT2C VUSs on the KMT2C DNAm-signature-derived SVM model (one individual with VUS c.14501T>C [p.Val4834Ala] was not available for testing). The SVM model resulted in classification of 5/21 VUSs as positive for the KMT2C DNAm signature. One out of 21 VUS was classified as intermediate (SVM score 0.25–0.5), and the rest were classified as negative on the KMT2C DNAm signature. These results supported re-classification of 6/21 variants as LP and 13/21 variants as benign (described below), while 2/21 variants remained in the VUS category (one additional VUS was unavailable for testing).

Next, we examined the basis of pathogenicity of the PAVs in this cohort. The details for each variant classification and the criteria applied utilizing various evidence are described in Table S4. All six variants reclassified as LP/P were classified as positive (or intermediate) on the DNAm signature. Out of the six VUSs reclassified as LP/P, two (de novo) variants were (re)interpreted as PTVs: duplication of exons 3–38, which is predicted to result in frameshift if present in tandem, so based on the DNAm signature classification, we concluded that the duplication most likely is indeed in tandem; one variant was initially reported from trio exome sequencing as de novo missense c.3499G>A (p.Asp1167Asn), located outside of known domains, but the signature analysis aided the reanalysis of the variant, which was later confirmed by Sanger sequencing as c.3499+1G>T (p.?). The remaining four LP PAVs (c.3233G>A [p.Cys1078Tyr], c.13229A>G [p.Asp4410Gly], c.13783C>T [p.Arg4595Cys], and c.14335C>G [p.Arg4779Gly]) are absent from gnomAD, predicted as pathogenic by in silico tools (REVEL⁴⁸ > 0.65 and AlphaMissense⁴⁹ > 0.56), and all are located in well-known functional domains of the protein (PHD6, ePHD2, FYRN, and SET domains, respectively). Moreover, these variants are predicted to disrupt the 3D structure of the respective domains (Figure S3) except for the c.13229A>G (p.Asp4410Gly) for which no reliable 3D structure or model was available.

In contrast, all 11/13 variants classified as benign are missense and are negative on the DNAm signature, the majority are not classified as pathogenic by in silico tools and are not predicted to affect protein 3D structure and/or located outside of the known functional domain’s (largely in disordered regions) (Table S4). Only three benign PAVs occurred de novo. Additionally, one splicing variant, c.1735+2dup, was inherited from father, predicted to result in inframe skipping of exon 12 that encodes a disordered region of the protein and was classified as negative by the DNAm signature, so the variant was classified as benign. Similarly, one benign partial KMT2C duplication (1–55 exons) with unknown inheritance was absent from population databases, but similar KMT2C 5′ partial duplications are common in the general population (Figure 1B), and it is also classified as negative on the DNAm signature.

Finally, three missense VUSs remained classified as VUSs due to conflicting or insufficient evidence. One of the missense variants (c.14501T>C [p.Val4834Ala]) is located in the SET domain, is predicted as pathogenic by in silico tools (REVEL = 0.8 and AlphaMissense = 0.99) and was predicted to disrupt the domain’s 3D structure (Table S2), but the sample was not available for DNAm testing. Two missense variants (c.9773A>C [p.His3258Pro] and c.13298C>T [p.Ala4433Val]) were classified negatively on the DNAm signature but had additional significant evidence for pathogenicity: both variants are de novo, absent in gnomAD, and predicted to be possibly disruptive by in silico tools and/or 3D protein analysis (Tables S2 and S4). These results show that the KMT2C DNAm signature can be used to reclassify most but not all VUSs in the gene and that PAVs disrupting functional domains are more likely to be P/LP.

KMT2C loss-of-function variants result in a phenotypically heterogeneous NDD syndrome

In total, we identified 81 individuals from 73 families with P/LP KMT2C variants. We gathered detailed clinical information on these individuals to understand the clinical spectrum of KMT2C-related NDD. Out of 81 individuals, six had a second molecular diagnosis due to a pathogenic variant in another NDD-associated gene. To understand the clinical consequences of P/LP KMT2C variants, we analyzed the clinical features of 75 individuals with P/LP KMT2C variants without any additional known P/LP variants in other NDD-associated genes (Tables 1 and S1).

Table 1.

Frequencies or distributions of clinical findings of KMT2C-related NDD

Clinical findinga	Frequency or distribution
Sex (males/females, n [%]) (n=75)	41 (54.7)/34 (45.3)
Age at last examination (years, m[IQR]) (n=71)	11.33 (5.17; 20)
Craniofacial dysmorphisms (n [%]) (n=70)	63 (90)
Antenatal and neonatal features
Abnormal pregnancy findings (n [%]) (n = 61)	14 (23)
Small/large for gestational age (n [%]) (n = 48)	9 (18.8)/2 (4.2)
Primary microcephaly/primary macrocephaly (n [%]) (n = 23)	3 (13)/1 (4.3)
Neonatal hypotonia (n [%]) (n = 52)	15 (28.8)
Neonatal feeding difficulties (n [%]) (n = 51)	25 (49)
Neurological and developmental features
Gross motor delay (n [%]) (n = 69)	60 (87)
Fine motor delay (n [%]) (n = 48)	38 (79.2)
Speech delay (n [%]) (n = 65)	52 (80)
Mutism (n [%]) (n = 47)	7 (14.9)
Developmental regression (n [%]) (n = 45)	6 (13.3)
Cognitive impairment (n [%]) (n = 69) Mild/moderate/severe (n [%])(n = 48)	59 (85.5) 24 (50)/12 (25)/12 (25)
Features or diagnosis of ASD (n [%]) (n = 61)	48 (78.9)
Features or diagnosis of ADHD (n [%]) (n = 51)	31 (60.8)
Other behavioral/psychiatric problems (n [%]) (n=61) Hetero/self-aggressive behavior (n [%]) Obsessive/compulsive behavior (n [%]) Other (n [%])	39 (63.9) 11 (18) 9 (14.8) 28 (45.9)
Seizure history (n [%]) (n = 66)	10 (15.2)
Hypotonia (n [%]) (n = 57)	19 (33.3)
CNS anomalies/abnormalities (n [%]) (n = 29)	10 (34.5)
Other neurological issues (n [%]) (n=61) Abnormal gait (n [%]) Other (n [%])	14 (22.9) 8 (13) 9 (15)
Endocrine anomalies (n [%]) (n=71) Short statureb (n [%]) Hypothyroidism (n [%])	40 (56.3) 39 (54.9) 3 (4)
Gastrointestinal/nutritional anomalies (n [%]) (n=43) Constipation (n [%]) Other (n [%])	23 (53.5) 6 (14) 21 (48.8)
Ophthalmological problems (n=60) Refractive error (n [%]) Strabismus (n [%]) Other (n [%])	32 (53.3) 20 (33.3) 12 (20) 4 (6.7)
Hearing impairment (n [%]) (n=58)	17 (29.3)
Musculoskeletal anomalies (n [%]) (n=64) Kyphosis and/or scoliosis (n [%]) Joint hypermobility (n [%]) Other (n [%])	32 (50) 10 (15.6) 5 (7.8) 25 (39.1)
Ectodermal anomalies (n [%]) (n=41) Hypertrichosis Other (n [%])	16 (39) 4 (9.8) 12 (29.3)
Immunological anomalies (n [%]) (n=43) Recurrent infections (n [%]) Other (n [%])	15 (34.8) 11 (25.6) 4 (9.3)
Respiratory anomalies (n [%]) (n=42) Central/obstructive sleep apnea (n [%]) Asthma (n [%]) Other (n [%])	14 (33.3) 7 (16.7) 6 (14.3) 2 (4.8)
Palate anomalies (n [%]) (n=37) High/narrow palate (n [%]) Cleft lip/palate (n [%])	12 (32.4) 11 (29.7) 1 (2.7)
Cardiovascular anomalies (n [%]) (n=54) Septal defects (n[%]) Conduction disorder (n[%]) Valvular anomalies (n [%]) Other (n [%])	13 (24.1) 6 (11.1) 6 (11.1) 4 (7.4) 3 (5.6)
Genitourinary anomalies (n [%]) (n=39) Uni/bilateral inguinal hernia (n [%]) Other (n [%])	8 (20.5) 2 (5.1) 7 (17.9)
Dental anomalies (n [%]) (n=70) Dental crowding (n [%]) Prominent upper incisors (n [%]) Other (n [%])	13 (18.8) 6 (8.6) 6 (8.6) 1 (1.4)

ADHD, attention deficit hyperactivity disorder; ASD, autism spectrum disorder; CNS, central nervous system; IQR, interquartile range; m, median; SD, standard deviation.

^aThe number of responders are detailed for every feature, and their frequencies/distributions were calculated according to that number. Individuals with another NDD were excluded.

^bAt any moment documented <−2SD.

In this cohort, 45.3% (n = 34) were females, and the median age at last examination was 11 years (with IQR 5.17; 20). The median weight at the last reported time was −1.01 SD (−1.94; −0.21), the median height was −1.56 SD (−2.32; −1.19), and the median head circumference (HC) was −0.44 SD (−1.65; 0.51). Importantly, short stature (≤−2 SD) was reported in ∼54% of individuals, and five of them have received recombinant human growth hormone (rhGH) with an apparently good response. Also, 9/53 (∼17%) individuals have displayed microcephaly (≤−2 SD) at least on one available measurement. These findings suggest KMT2C-related NDD is mainly an undergrowth condition.

Neurodevelopmental problems are the most common features in this cohort. Although gross motor delay was described in 87% of individuals, all individuals older than 3.5 years of age had achieved independent walking. Speech delay was reported in 80% of individuals, and ∼15% developed mutism at some point in their lifetime. Developmental regression was reported in six individuals. Intellectual disability was reported in ∼86% of individuals with half being classified as having moderate or severe intellectual disability. Importantly, a proportion of individuals had normal cognitive abilities (∼14%; all with de novo variants) and some of them were tested due to non-NDD phenotypes (e.g., short stature, seizures). However, we cannot exclude mosaicism in these cases. Furthermore, behavioral and psychiatric problems were reported in most individuals: features or diagnosis of ASD or attention deficit hyperactivity disorder (ADHD) were reported in ∼79% and 61% of individuals, respectively; aggressive (both hetero- and self-) behavior was reported in 18% and obsessive-compulsive behavior in ∼15%.

Abnormalities affecting the central nervous system (∼34.5%; e.g., ventriculomegaly, white-matter anomalies, syringomyelia) and the cardiovascular system (∼24%; e.g., septal defects, valvular anomalies) were frequent. Refractive errors (33%) and hearing loss (∼29%) were the most frequent sensory system problems. Recurrent infections (∼26%), central/obstructive sleep apnea (∼17%), seizures (∼15%), kyphosis and/or scoliosis (∼16%), and constipation (∼16%) were some of the other most significant and frequently encountered medical issues (Table 1). Feeding difficulty was the most frequent neonatal finding (49%) followed by neonatal hypotonia (∼29%).

Next, we compared the frequencies of clinical features between male and female individuals, type of variants (PAVs vs. PTVs), and their inheritance (inherited vs. de novo). None of the features were significant after correction for multiple testing (p_adj > 0.05).

Craniofacial dysmorphisms were described in 90% of individuals with P/LP KMT2C variants in our cohort. Photographs of 34 individuals were available in this cohort, with 13 consenting for publication (Figure 3). We used these photographs to analyze the facial features of individuals with KMT2C-related NDD. The most frequent findings were frontal bossing/prominent-broad forehead, thick/prominent eyebrows, synophrys, down-slanted palpebral fissures, deep-set eyes, epicanthus, hypertelorism, midface retrusion, anteverted nares, thin vermillion of the upper lip, micrognathia, low-set ears, thick ear helices, and posteriorly rotated ears. Although all individuals had craniofacial dysmorphism, overall, we did not identify a clearly recognizable KMT2C facial gestalt. Next, to objectively evaluate whether the KMT2C-related NDD has a specific facial gestalt, we used PhenoScore and compared the facial photographs of individuals with the KMT2C-related NDD with facial photographs of individuals in a general intellectual disability cohort. PhenoScore identified a specific KMT2C-related facial gestalt (area under the curve [AUC] = 0.91, p < 0.001) (Figure S2), indicating that the facial features the 29 individuals with an LP/P KMT2C variant for whom an age-, sex-, and ethnicity-matched NDD control was available are distinguishable from the general NDD population. The PhenoScore identified periorbital and nasal regions as the most different (Figure S2), which fit the most common dysmorphic features described above.

Figure 3.

Photos of individuals with the KMT2C-related NDD

y.o., years old—the age of the individuals.

Collectively, these observations show that KMT2C-related NDD is clinically a highly heterogeneous disorder characterized by intellectual disability, short stature, congenital anomalies, recurrent infections, and craniofacial dysmorphism.

KMT2C-related NDD is distinct from Kleefstra and Kabuki syndromes

OMIM designates the KMT2C-related NDD as “Kleefstra syndrome 2” (MIM: 617768), which may be interpreted that the EHMT1- and KMT2C-related disorders have significant clinical overlap. However, this overlap has not been examined previously. Also, KMT2C and KMT2D are part of the same gene family, but their overlap has not been examined previously. For this, firstly, we compared the facial features of individuals with KMT2C-related NDD to the photographs of 24 and 10 matched individuals with Kleefstra, and Kabuki type 1 syndromes, respectively. PhenoScore revealed significant differences between the KMT2C-related NDD facial gestalt and the facial gestalt of Kleefstra syndrome (AUC = 1, p < 0.001) and Kabuki syndrome type 1 (AUC = 0.8, p = 0.04).

DNAm signatures can also be used to delineate epigenetically distinct disorders.^50,51 Pathogenic loss-of-function variants in EHMT1⁵² or KMT2D⁶ result in unique DNAm signatures.^51,53 We analyzed DNA samples from six individuals with Kleefstra syndrome and six individuals with Kabuki 1 syndrome on the KMT2C DNAm signature. Samples from Kleefstra and Kabuki type 1 syndrome were classified negatively by the SVM (Figure 2D; Table S5) on the KMT2C DNAm signature. Similarly, all tested KMT2C-related NDD individuals were classified negatively on the Kleefstra and Kabuki syndrome signatures (Figure 2E), highlighting the DNAm differences among the three conditions. These findings confirm that the KMT2C-related NDD is clinically and epigenetically distinct from both Kleefstra and Kabuki syndromes.

Discussion

In this study, we have demonstrated that heterozygous loss-of-function KMT2C variants result in a disorder that is characterized by variable combinations of developmental delay, intellectual disability, short stature, congenital anomalies, craniofacial dysmorphism, recurrent infections, and a moderate-strength DNAm signature. We also demonstrate that the KMT2C-related NDD is clinically and epigenetically distinct from (EHMT1-related) Kleefstra syndrome type 1 and (KMT2D-related) Kabuki type 1 syndromes.

Our data show that despite the complexities of the transcripts encoded by KMT2C, the PTVs in the gene have similar clinical or epigenetic consequences irrespective of their location in the gene. This observation suggests four possible explanations: (1) the canonical isoform is biologically and disease relevant in spite of its apparent low level of expression in tissues, (2) the developmental expression profile of KMT2C may be different from its known adult expression profile, (3) the KMT2C transcript expression pattern may not yet be fully known, or (4) the regulatory consequences of loss-of-function variants in the longer transcripts may have complex consequences on the transcriptional network. Additionally, we cannot exclude that the low expression of the canonical transcript in GTEx is a technical artifact of the short-read-based RNA sequencing, displaying 3′ bias and short-isoform preferences.⁴² Our analysis also shows that KMT2C is located in a hot-spot for both benign and pathogenic structural rearrangements, and therefore, CNVs in this gene should be interpreted with caution. In our study, we also showed that rare missense variants that affect conserved residues and disrupt the 3D structure of the PHD Zn fingers, the FYR domain, or the catalytic SET also cause the same disorder. Interestingly, pathogenic Kabuki syndrome type 1-causing missense variants in KMT2D, a paralog of KMT2C, cluster significantly in several similar functional domains.⁵⁴ These findings collectively indicate haploinsufficiency and functional haploinsufficiency as likely mechanisms that underlie KMT2C-related NDD.

We also identify a moderate-effect DNAm signature for KMT2C-NDD with current sensitivity of ∼82%. To identify a reliable signature, we had to utilize a relatively large sample size cohort (of different ages and sex) as initial signatures derived from smaller cohort sizes lacked sensitivity and specificity. It is possible that variable expressivity is typical not only for the clinical features, but also for the DNAm signature. It has been shown previously that the level of DNAm changes correlate with the age of onset of dystonia in the KMT2B-related dystonia⁵⁵ (MIM: 617284), and the Weaver-syndrome (MIM: 277590) signature was found to be weaker in mildly affected individuals with a pathogenic EZH2 (MIM: 601573) variant even within a single family.¹⁸ Therefore, it is possible that mildly affected individuals can be classified as intermediate or negative on the DNAm signature despite the presence of pathogenic variants. For example, one mildly affected individual with a pathogenic KMT2C variant (exons 36–37 deletions) classified as negative by the DNAm signature is the mother of three affected children. This individual is mildly affected, which might explain the DNAm classification results. Unfortunately, DNA from the children was not available for testing. However, the second individual with pathogenic variant c.13107_13108dup (p.Thr4370Argfs^∗11) was also classified negatively despite having typical KMT2C-related NDD clinical features. However, we were not able to exclude mosaicism in either case, which may explain the negative classification results. In the future, larger studies will be required to further elucidate the biological basis of such correlations, which may have prognostic and management implications.

A large proportion of the identified KMT2C signature’s CpG sites (8%) mapped to WT1 CpG islands, which are hypomethylated in KMT2C-related NDD individuals. Hypomethylation of CpG islands, especially those located in the promoter of a gene, is usually associated with upregulation of gene expression,⁵⁶ which was previously shown for WT1 in vitro.⁵⁷ WT1 is not a known interactor of KMT2C, so it is unclear whether the hypomethylation at the WT1 CpG islands is a primary effect of KMT2C disruption or secondary to disrupted interactions of multiple genes. Loss-of-function or dominant-negative WT1 variants are associated with Wilms tumor (MIM: 194070) or Denys-Drash syndrome (MIM: 194080), respectively, but to date, there is no human phenotype associated with WT1 overexpression. Until now, individuals with KMT2C-related NDD have not displayed any tumors or anomalies of sex development that are reported for the WT1-related disorders. Recently, a role for WT1 in brain and neuron development was also described.^58,59,60 Mariotinni et al.,⁶¹ demonstrated that WT1 is important for synaptic plasticity and learning in mice and that it functions as a long-term memory suppressor. Additionally, they showed that Wt1 overexpression causes a reduction of memory retention.⁶¹ In another study, Ji et al. showed that Wt1 brain-specific loss in mice resulted in depressive-like behavior, but the effects of overexpression were not investigated.⁶⁰ At this time, the relevance of WT1 methylation changes to the phenotype of KMT2C-related NDD is not known.

Several neurodevelopmental phenotypes were present in the majority of the individuals in the study cohort, including developmental delay, intellectual disability (mostly mild, but ranging from borderline to severe), ASD, and ADHD. Additionally, the individuals often have other psychiatric issues, including hetero- or self-aggressive behavior, obsessive-compulsive behavior, and selective mutism. In fact, pathogenic or de novo KMT2C variants have also been found to be enriched in individuals with schizophrenia,⁶² bipolar disorder,⁶³ and other psychiatric disorders.⁶⁴ This suggests that KMT2C-related NDD is associated with multiple, overlapping neurodevelopmental and psychiatric presentations.

Individuals with KMT2C-related NDD present with a broad spectrum of clinical features, but the expressivity of the features is highly variable among individuals. For example, ∼14% of the individuals in the entire cohort did not have intellectual or learning disabilities. The variable expressivity likely explains the relatively large proportion (11/75, ∼15%) of inherited cases in this cohort, as most of the parents with pathogenic variants were mildly affected.

Individuals with the KMT2C-related NDD had common dysmorphic features, including frontal bossing or prominent forehead, down-slanted palpebral fissures, deep-set eyes, and hypertelorism among others. These features are mild and non-specific, and therefore, the facial gestalt was unrecognizable to clinicians, unlike Kleefstra and Kabuki 1 syndromes. However, using PhenoScore,³⁸ we were able to show that the KMT2C-related NDD had a typical facial gestalt that is significantly different from Kleefstra and Kabuki 1 syndromes.

KMT2C-related NDD, Kleefstra, and Kabuki 1 syndromes are clearly different conditions despite some clinical and molecular overlap, but it is currently impossible to objectively compare the frequencies of specific features among these conditions head to head. Kleefstra and Kabuki syndromes are clinically well known and well recognized, and most of the published individuals are diagnosed through a phenotype-first approach.^8,65 This results in biased results with only the most typical affected individuals being initially described. However, with the recent widespread application of exome or genome sequencing in clinical diagnostics, it is possible to also identify non-specific and/or mildly affected individuals,^66,67 as well as novel morbid entities within the same gene,⁵⁰ including KMT2D.⁶⁸ KMT2C-related NDD is not clinically recognizable, so all individuals were diagnosed through a genotype-first approach, and this study represents a clinical expansion and delineation of this condition. Therefore, studies on the phenotype of Kleefstra and Kabuki 1 syndromes diagnosed through a genotype-first approach would be necessary to better understand their clinical and molecular spectrum, as well as the differences among the three conditions.

Although KMT2C-related NDD and Kleefstra and Kabuki 1 syndromes are caused by haploinsufficiency of methyltransferases, where PTVs are the most frequent type of variant, the proportion of inherited variants is considerably different among them. While Kleefstra and Kabuki 1 syndrome mostly occur by de novo loss-of-function variants, with only a handful of inherited cases described so far,^69,70 ∼15% of KMT2C-related NDD individuals inherited their variants from similarly affected parents. Albeit, this may be explained by a biased approach to diagnosis; our study also showed that the expressivity of the KMT2C-related NDD is highly variable, and individuals with few manifestations can be expected.

Several individuals in our study were primarily investigated due to short stature—a feature present in more than a half of the KMT2C-related NDD individuals reported in this study. Short stature is also common in Kleefstra and Kabuki 1 syndromes.^8,65 The causes of short stature in these syndromes are often unknown but likely complex.⁶⁵ However, short stature in individuals with monogenic NDDs is rarely investigated in-depth in routine clinical practice as it is usually considered to be part of the syndrome. While it is known that growth hormone deficiency is present in approximately a third of individuals with Kabuki 1 syndrome,⁷¹ this can partially explain the presence of short stature. However, treatment with rhGH has proven to be highly effective and safe,⁷² and therefore, these individuals should be screened for this deficiency. In this cohort, 40/75 individuals with endocrine anomalies had proven short stature, and 6/40 had received rhGH with apparently good results. This should prompt further studies on the prevalence of GH deficiency and the utility of rhGH therapy also for individuals with KMT2C-related NDD and short stature.

While various psychiatric disorders are seen in KMT2C-related NDD, this is not frequent in Kabuki 1 syndrome.⁷³ Also, although psychiatric disorders such as schizophrenia⁷⁴ and developmental regression^75,76,77 have been frequently reported in Kleefstra syndrome, nearly all individuals with Kleefstra syndrome manifest intellectual disability that ranges from moderate to severe,^8,77 which contrasts with our finding that ∼15% of individuals with KMT2C-related NDD demonstrate no cognitive impairment. Again, albeit this may be explained by a bias in our diagnostic approach, our study shows that the penetrance and severity of this finding is consistently reduced in the KMT2C-related NDD.

In summary, pathogenic variants in KMT2C result in a distinct syndromic neurodevelopmental disorder, which is distinct from Kleefstra and Kabuki 1 syndromes. Therefore, the OMIM designation of KMT2C-related NDD as Kleefstra syndrome 2 should be reconsidered, as it may be misleading to clinicians and patients’ families.

Data and code availability

Clinical details of recruited individuals with pathogenic KMT2C variants are provided in the Table S1. Details of recruited individuals with benign or VUS KMT2C variants are provided in the Table S2. The variants and their interpretations were submitted to the ClinVar database (ClinVar: SCV005044911-SCV005044983 and SCV005044991-SCV005045000).

The KMT2C DNAm signature is available in supplementary material (Table S3). The raw datasets supporting the current study have not been deposited in a public repository due to institutional ethics restrictions. All software and R packages used in the study are publicly available as described in the methods section. The KMT2C methylation classifier will be made available through EpigenCentral at https://epigen.ccm.sickkids.ca.

Published: July 15, 2024

Web resources

EpigenCentral, https://epigen.ccm.sickkids.ca