DNA methylation signature associated with Bohring-Opitz syndrome: a new tool for functional classification of variants in ASXL genes

Abstract

The additional sex combs-like (ASXL) gene family-encoded by ASXL1, ASXL2, and ASXL3-is crucial for mammalian development. Pathogenic variants in the ASXL gene family are associated with three phenotypically distinct neurodevelopmental syndromes. Our previous work has shown that syndromic conditions caused by pathogenic variants in epigenetic regulatory genes show consistent patterns of genome-wide DNA methylation (DNAm) alterations, i.e., DNAm signatures in peripheral blood. Given the role of ASXL1 in chromatin modification, we hypothesized that pathogenic ASXL1 variants underlying Bohring-Opitz syndrome (BOS) have a unique DNAm signature. We profiled whole-blood DNAm for 17 ASXL1 variants, and 35 sex- and age-matched typically developing individuals, using Illumina's Infinium EPIC array. We identified 763 differentially methylated CpG sites in individuals with BOS. Differentially methylated sites overlapped 323 unique genes, including HOXA5 and HOXB4, supporting the functional relevance of DNAm signatures. We used a machine-learning classification model based on the BOS DNAm signature to classify variants of uncertain significance in ASXL1, as well as pathogenic ASXL2 and ASXL3 variants. The DNAm profile of one individual with the ASXL2 variant was BOS-like, whereas the DNAm profiles of three individuals with ASXL3 variants were control-like. We also used Horvath's epigenetic clock, which showed acceleration in DNAm age in individuals with pathogenic ASXL1 variants, and the individual with the pathogenic ASXL2 variant, but not in individuals with ASXL3 variants. These studies enhance our understanding of the epigenetic dysregulation underpinning ASXL gene family-associated syndromes.

Fig. 1. Genomic Location of ASXL1 variants.

Schematic representation of the ASXL1 protein (GenBank: ASXL1; NM_015338.6; GRCh37), its functional domains, and variants used in this study. Exon structure, based on GenBank: NM_015338.6, is provided by dashed lines. Red, HB1, ASXL, restriction endonuclease HTH domain (HARE-HTH, 11–83); purple, Asx homology domain (ASXH, 236–359); green, C-terminal plant homeodomain (PHD, 1506–1539). The N-terminal HARE-HTH domain is DNA binding and with the ASXH domain are required for interaction with BAP1 and NCOA1. The c-terminal PHD is required for interaction with nuclear receptors. The map was generated using ProteinPaint [31].

Fig. 2. Individuals with ASXL1 variants exhibit altered epigenetic aging.

A Box plot comparing “DNA methylation age” (blue) derived from the Illumina 850 K data and reported chronological age (red), on the y-axis. Groups are indicated on the x-axis and include typically developing controls (n = 35), individuals with ASXL1 variants (n = 17), individual with ASXL2 variant (n = 1), and individuals with ASXL3 variants (n = 3). Each individual observation is plotted as a circle. To assess whether the mean difference between DNAm age and chronological age is statistically significant within each group we used a paired Wilcoxon test (*p-value < 0.05). (B) Box plot of epigenetic age acceleration (y-axis) obtained by subtracting the chronological age from the estimated DNAm age for each individual. To assess whether mean epigenetic age acceleration estimates are significantly different between controls and individuals carrying variants in ASXL genes we used a Mann-Whitney U-Test, except for ASXL2 with a n = 1.

Fig. 3. Loss-of-function variants in ASXL1 are associated with a distinct DNAm signature.

A Principal component analysis (PCA) and (B) heatmap showing clustering of the BOS discovery cohort (n = 8; yellow) and control discovery cohort (n = 26; grey) using DNAm values at the 763 CpG sites identified in the BOS specific DNAm signature. The heatmap color gradient indicates the normalized DNAm value ranging from −2.0 (blue) to 2.0 (yellow). Euclidean distance metric is used in the heatmap clustering dendrograms.

Fig. 4. Classification of samples using machine learning models based on the BOS DNAm signature.

Sample groups were scored using the BOS support vector machine (SVM) model. The x-axis groups each cohort, and the y-axis shows the probability score. BOS validation subject (n = 6) had high probability scores demonstrating 100% sensitivity of the model. Whereas validation control subjects (n = 101) all had low scores demonstrating 100% specificity of the model. ASXL1 missense variants (n = 3) and ASXL3 truncating variants (n = 3) scored low similar to controls, whereas the truncating ASXL2 variant (n = 1) scored high similar to the BOS validation group. Lastly, individuals with Sotos, Weaver, and Kabuki syndromes, caused by pathogenic variants in the chromatin-modifying genes NSD1, EZH2, and KMT2D respectively, all scored low further demonstrating 100% specificity of the model. Horizontal line represents threshold for classifying samples as case-like (above line) or control-like (below line).