Genome-Wide DNA Methylation Profiling Solves Uncertainty in Classifying NSD1 Variants

Abstract

Background: Inactivating NSD1 mutations causing Sotos syndrome have been previously associated with a specific genome-wide DNA methylation (DNAm) pattern. Sotos syndrome is characterized by phenotypic overlap with other overgrowth syndromes, and a definite diagnosis might not be easily reached due to the high prevalence of variants of unknown significance (VoUS) that are identified in patients with a suggestive phenotype.

Objective: we performed microarray DNAm profiling in a set of 11 individuals with a clinical suspicion of Sotos syndrome and carrying an NSD1 VoUS or previously unreported variants to solve uncertainty in defining pathogenicity of the observed variants. The impact of the training cohort size on sensitivity and prediction confidence of the classifier was assessed.

Results: The Sotos syndrome-specific DNAm signature was validated in six individuals with a clinical diagnosis of Sotos syndrome and carrying bona fide pathogenic NSD1 variants. Applying this approach to the remaining 11 individuals with NSD1 variants, we succeeded in confirming pathogenicity in eight subjects and excluding the diagnosis of Sotos syndrome in three. The sensitivity and prediction confidence of the classifier based on the different sizes of the training sets did not show substantial differences, though the overall performance was improved by using a data balancing strategy.

Conclusions: The present approach solved uncertainty in cases with NDS1 VoUS, further demonstrating the clinical utility of DNAm profiling.

Keywords: DNA methylation; NSD1; Sotos syndrome; VoUS validation; differential diagnosis; genomic variant classification; overgrowth.

Figure 1

Facies of three subjects with clinical diagnosis of Sotos syndrome carrying functionally unclassified NSD1 variants. Clinical features of S10 (1 year), S12 (2 years) and S14 (6 months, left panels; 1 year and 3 months, right panels) showing typical Sotos syndrome characteristics. All subjects show macrocephaly, dolichocephaly, broad and prominent forehead, sparse fronto-temporal hair, downslanted palpebral fissures, long and narrow face, and a long/pointed chin.

Figure 2

Mutational landscape for NSD1. MutLand plot for NSD1 (NM_022455.4) [9], showing exons, functional domains and variants reported in ClinVar. Location of the variants included in the present study are also shown, as per different ACMG classification (blue, VoUS; red, likely pathogenic). Positions of the missense and LoF NSD1 variants reported in ClinVar known to be either pathogenic/likely pathogenic (PLP) or benign/likely benign (BLB) are shown (top diagrams), whereas VoUS or variants having conflicting interpretation (CI) reported by ClinVar, either missense or LoF, are shown in the bottom diagram. The cartoons highlight the regions for which MutScore detects significant PLP clustering (orange horizontal bars). A diagram reporting the functional domains of the protein (PHD, yellow; PWWP, red; SET, light green), and plotting the MutScore referred for the entire protein sequence is also shown (third plot from the top).

Figure 3

DNAm array analyses. (A) DNAm profiles in individuals carrying NSD1 variants versus a control cohort. Taking advantage of DNAm levels from the established episignature, MDS plot was used to test/validate sample clustering with respect to the Sotos syndrome training cohort (red, solid circles) and a control group including 295 healthy individuals and subjects with rare neurodevelopmental disorders (blue, solid circles). Tested VoUS are depicted in orange (solid circles), while control subset used for testing are in grey (solid circles). (B) Plot for SVM probability scores from the developed ML-based classifier, showing classification results for the tested samples (VoUS, n = 8, orange solid circles; validation samples, n = 1, red open circles), and the in-house control cohort (grey = testing).