DNA Methylation Signature for EZH2 Functionally Classifies Sequence Variants in Three PRC2 Complex Genes

Abstract

Weaver syndrome (WS), an overgrowth/intellectual disability syndrome (OGID), is caused by pathogenic variants in the histone methyltransferase EZH2, which encodes a core component of the Polycomb repressive complex-2 (PRC2). Using genome-wide DNA methylation (DNAm) data for 187 individuals with OGID and 969 control subjects, we show that pathogenic variants in EZH2 generate a highly specific and sensitive DNAm signature reflecting the phenotype of WS. This signature can be used to distinguish loss-of-function from gain-of-function missense variants and to detect somatic mosaicism. We also show that the signature can accurately classify sequence variants in EED and SUZ12, which encode two other core components of PRC2, and predict the presence of pathogenic variants in undiagnosed individuals with OGID. The discovery of a functionally relevant signature with utility for diagnostic classification of sequence variants in EZH2, EED, and SUZ12 supports the emerging paradigm shift for implementation of DNAm signatures into diagnostics and translational research.

Keywords: DNA methylation signature; EED; SUZ12; intellectual disability; overgrowth syndromes.

Figure 1

Comprehensive Visualization of EZH2 Sequence Variants using ProteinPaint Schematic representation of EZH2 sequence variants included in the study. Each distinct variant in EZH2 is represented by a disc sized in proportion to the number of samples and filled with the color representing its class based on the legend. Missense variants which constitute the large proportion of variants in EZH2 are colored in red, nonsense variants in blue, and indels in green. Sequence variants are positioned by their amino acid coordinates based on EZH2 (GenBank: NM_004456.4, hg19). The dotted vertical lines inside the protein delineate the boundaries of coding exons and the filled colors within the protein correspond to known protein domains.

Figure 2

EZH2-Specific DNAm Signature Principal components analysis (PCA) plot (A) and corresponding hierarchical clustering (B) (Eucledian distance metrics) and representative heatmap of 31 samples (n = 8 WS; n = 23 controls) using the differentially methylated CpG sites comprising the EZH2-specific DNAm signature (229 CpG sites). In both (A) and (B), samples labeled with red represent Weaver syndrome, blue samples are controls. On the heatmap, yellow indicates high DNAm and blue indicates low DNAm. For the heatmap, data are normalized for visualization (mean = 0, variance = 1).

Figure 3

Testing the Sensitivity and Specificity of the EZH2-Specific Signature (A) Plot representing the median-methylation profiles of WS-affected individuals (y axis) and control subjects (x axis) using the EZH2 DNAm signature. The dashed line is set to represent the decision boundary for which individuals above the dashed lines have DNAm profiles more similar to EZH2 signature and below the dashed line have DNAm profiles more similar to control subjects. A set of independent WS-affected individuals (validation cohort, purple circles, n = 8) as well as a WS-affected family (light orange circles, n = 5 affected members) with EZH2 pathogenic variants were classified as “WS” (i.e., all individuals classified as more similar to the EZH2 signature than control subjects) indicating high accuracy of the EZH2 DNAm signature. The specificity of EZH2 signature was estimated on an independent control validation set of 148 control samples (green crossed boxes); all subjects classified as more similar to control subjects (specificity 100%). (B) Performance of the EZH2 signature on data generated on 450k array including overlapping WS-affected subjects from the discovery cohort n = 7 (red circles), controls n = 80 (blue squares), and GEO controls n = 718 (brown squares) generated on 450k array. All control subjects had DNAm profiles more similar to the control profile and were therefore classified as “not-WS.” WS, Weaver syndrome.

Figure 4

Testing the Ability of the EZH2-Specific Signature to Classify EZH2 Variants (A) Plot representing the median-methylation profiles of WS-affected (y axis) and control subjects (x axis) using the EZH2 DNAm signature. A set of independent individuals with EZH2 sequence variants (pink squares, n = 19) were classified using the EZH2 signature. Of the 19 variants, ten classified as more similar to the EZH2 DNAm profile than controls. The remaining nine variants classified as more similar to the control profile. (B) Plot representing the Support Vector machine (SVM) scores (y axis). The SVM prediction model was used to predict pathogenicity of EZH2 variants based on the DNAm signature. All ten variants predicted as pathogenic in (A) had also very high SVM scores > 70% and the remaining nine variants had very low SVM scores < 20% except one variant with an SVM score of 49%. Blue arrow represents sample MDL#67845 (p.Ser669Asn). Orange arrow represents sample S126694 (EZH2_c.2196-2_2211dupAGATACAGCCAGGCTGAT). Green arrow represents sample A1646 (p.Ala738Thr). WS, Weaver syndrome.

Figure 5

Gain-of-Function Variant in EZH2 Have Opposite DNA Methylation Profile at the EZH2 Signature (A) Heatmap showing the hierarchical clustering of the DNAm profile of WS individuals (n = 8, red) with LoF (hypomorphic) variants in EZH2, controls (n = 23, blue) and EZH2 GoF variant (p.Ala738Thr; pink) using the EZH2 signature. On the heatmap, yellow indicates high DNAm and blue indicates low DNAm. The subject with an EZH2 variant in pink display opposite DNAm profile when compared to WS DNAm profiles. For the heatmap, data are normalized for visualization (mean = 0, variance = 1). (B) Enzymatic activity of EZH2 GoF variant using an in vitro luminescence assay. Mutant EZH2 (p.Ala738Thr) pre-assembled into PRC2 showed increased EZH2-mediated H3K27 methylation activity. WS, Weaver syndrome; GoF, gain of function; LoF, loss of function.

Figure 6

Testing the Utility of the EZH2 Signature in Classifying Sequence Variants in Other Components of the PRC2 Complex Using the EZH2 signature, we compared the DNAm profiles of three subjects with EED sequence variants (yellow triangles) and those with SUZ12 variants (green diamonds) to the DNAm profiles of controls (blue crossed boxes) and EZH2 pathogenic variants (red circles). All three subjects with pathogenic variants in EED had DNAm profiles more similar to the EZH2 profile than control subjects. Two subjects with SUZ12 pathogenic variants also classified with Weaver syndrome and the three remaining SUZ12 variants showed DNAm profiles more similar to controls.

Figure 7

Classification of Subjects with Syndromic Overgrowth using EZH2 Signature Plot representing samples with sequence variants in epigenes associated with other overgrowth syndromes. These included subjects with: NSD1 pathogenic variants associated with Sotos syndrome (n = 49, GEO: GSE74432), DNMT3A pathogenic variants associated with Tatton-Brown Rahman syndrome (n = 5; GEO: GSE128801), and CHD8 pathogenic variants associated with macrocephaly and susceptibility to autism (n = 10; GEO: GSE113967). These data were compared to seven WS-affected subjects from the discovery cohort and five test individuals with pathogenic mutations in EZH2 which were run on the Illumina 450k. All overgrowth syndrome individuals had DNAm profiles more similar to control subjects and distinguishable from WS.

Figure 8

Testing the Ability of the EZH2 Signature in Classifying Undiagnosed OGID-Affected Subjects Based on Their DNAm Profiles OGID-affected subjects included in this analysis were previously tested negative for targeted mutations screening in NSD1 and EZH2. Out of the 73 subjects with OGID (brown triangles), we identified that most had DNAm profiles similar to control subjects (blue squares). Interestingly, we identified two subjects with DNAm profiles more similar to the EZH2 profile than control samples. OGID, overgrowth and intellectual disability.