Diagnostic Utility of Genome-wide DNA Methylation Testing in Genetically Unsolved Individuals with Suspected Hereditary Conditions

Abstract

Conventional genetic testing of individuals with neurodevelopmental presentations and congenital anomalies (ND/CAs), i.e., the analysis of sequence and copy number variants, leaves a substantial proportion of them unexplained. Some of these cases have been shown to result from DNA methylation defects at a single locus (epi-variants), while others can exhibit syndrome-specific DNA methylation changes across multiple loci (epi-signatures). Here, we investigate the clinical diagnostic utility of genome-wide DNA methylation analysis of peripheral blood in unresolved ND/CAs. We generate a computational model enabling concurrent detection of 14 syndromes using DNA methylation data with full accuracy. We demonstrate the ability of this model in resolving 67 individuals with uncertain clinical diagnoses, some of whom had variants of unknown clinical significance (VUS) in the related genes. We show that the provisional diagnoses can be ruled out in many of the case subjects, some of whom are shown by our model to have other diseases initially not considered. By applying this model to a cohort of 965 ND/CA-affected subjects without a previous diagnostic assumption and a separate assessment of rare epi-variants in this cohort, we identify 15 case subjects with syndromic Mendelian disorders, 12 case subjects with imprinting and trinucleotide repeat expansion disorders, as well as 106 case subjects with rare epi-variants, a portion of which involved genes clinically or functionally linked to the subjects' phenotypes. This study demonstrates that genomic DNA methylation analysis can facilitate the molecular diagnosis of unresolved clinical cases and highlights the potential value of epigenomic testing in the routine clinical assessment of ND/CAs.

Keywords: DNA methylation; clinical epigenomic testing; congenital anomaly; epi-signature; epi-variant; imprinting; multiclass prediction; neurodevelopmental syndrome; nucleotide expansion repeat; unresolved case; variant of unknown significance.

Figure 1

Dimension Reduction of DNA Methylation Data in Constitutional Syndromes Using t-distributed stochastic neighbor embedding (t-SNE), we reduced the dimension of the DNA methylation data of all of the prioritized features (856 CpGs) in all of the individuals with the constitutional syndromes in the study. The data were reduced to two dimensions as presented on the x and y axes. Every point represents one subject. Subjects in the training cohort who were used for feature selection are shown in lighter colors, and those from the testing cohort who were not used for the prioritization of features are shown in darker colors. The plot shows that every disease generates a distinct cluster from the others and the samples in the testing cohort cluster within the correct disease group. BAFopathies, regardless of the subtype, generate one cluster, while those with ADNP syndrome, depending on the mutation coordinates, generate two distinct groups (ADNP-1 and ADNP-2). The observed pattern could not be explained by the experimental batch structure of the data. Abbreviations: CdLS, Cornelia de Lange syndrome; CJS, Claes-Jensen syndrome; FHS, Floating-Harbor syndrome; GTPTS, genitopatellar syndrome.

Figure 2

Concurrent Classification of Constitutional Syndromes using DNA Methylation Data A multi-class SVM classifier generates 16 scores ranging from 0 to 1 for every given individual (y axes), as the level of confidence for having any of the 16 phenotypes for which the model is trained for (x axes). The first 15 classes represent different syndromes and the last one indicates a state that the DNA methylation profiling is not indicative of any of the disorders in the model (healthy control subjects or other diseases). The figure shows trials performed for individuals (represented with hollow points) known to have one of the constitutional syndromes indicated on the x axes, from both the training (blue) and testing (red) cohorts (Table S1). Every panel illustrates trials for a group of subjects with a distinct phenotype as indicated on top of the panel. The first 15 panels show that every affected subject has received the highest score in the correct disease category. Subjects from the testing cohort (red) have also received scores similar to those in the training cohort (blue) with the highest scores in the correct categories. The last panel represents a total of 3,307 subjects composed of 2,880 healthy control subjects and 442 case subjects with confirmed diagnoses of constitutional disorders other than those in the model, including imprinting defect disorders (Angelman syndrome [n = 14], Prader-Willi syndrome [n = 7], Silver-Russell syndrome [n = 64], and Beckwith-Wiedemann syndrome [n = 9]), Weaver syndrome (n = 7), Saethre-Chotzen syndrome (n = 25), Rett syndrome (n = 28), Coffin-Lowry syndrome (n = 11), RASopathies (Noonan syndrome [n = 69], cardiofaciocutaneous syndrome [n = 15], and LEOPARD syndrome [n = 2]), fragile X syndrome (n = 50), and autism spectrum disorders (n = 141). All of these subjects are correctly classified into the final category. Due to small sample sizes, no testing cohort is available for subjects with ADCADN and GTPTS. Abbreviations: CdLS, Cornelia de Lange syndrome; CJS, Claes-Jensen syndrome; FHS, Floating-Harbor syndrome; GTPTS, genitopatellar syndrome.

Figure 3

Classification of Uncertain Clinical Case Subjects The model presented in Figure 2 was used to assess subjects with uncertain clinical/molecular diagnoses. The above figure shows three groups of uncertain case subjects for at least one of whom in every category the prediction was found to be different from what was initially suspected. The first panel shows 11 subjects suspected to have CHARGE syndrome, out of which 5 are classified to have CHARGE syndrome and another 5 are predicted to have none of the syndromes in the model. The final subject (shown in red), however, has received the highest score for the ADNP-1 category and low scores for all others. The second panel represents trials performed for those suspected of having Claes-Jensen syndrome (n = 6) among whom 4 case subjects of CJS and 1 case subject of Sotos syndrome (red) were identified. Similarly, in the last panel, among 26 subjects being assessed for Kabuki syndrome, 1 case of Claes-Jensen syndrome is found in addition to 8 case subjects with Kabuki syndrome (details in Table S5). Abbreviations: CdLS, Cornelia de Lange syndrome; CJS, Claes-Jensen syndrome; FHS, Floating-Harbor syndrome; GTPTS, genitopatellar syndrome.

Figure 4

Examination of DNA Methylation Profiles in Unresolved Case Subjects Screening of 965 unresolved ND/CA-affected case subjects using the model presented in Figure 2 identified a total of 15 subjects (Table S6) as potential cases of 10 of the syndromes in the model. Using the CpG probes specific to every condition, we compared the DNA methylation profiles of the identified subjects to a few representative diagnosed case subjects and healthy control subjects. The analysis was done using hierarchical clustering by Ward’s method on Euclidian distance. In every panel above, rows represent the CpG probes and columns represent the subjects. The heatmap panes indicate the phenotype. Blue, healthy control subjects; green, individuals with the confirmed diagnosis of the syndromes; red, the candidate subject(s) under assessment. The heatmap color scale from green to red represents the range of the methylation levels (beta values) between 0 and 1. The analysis indicates that every candidate case subject presents a DNA methylation pattern consistent with the predicted phenotype, i.e., ADNP (A), ATRX (B), BAFopathies (C), CHARGE (D), Cornelia de Lang (E), Claes-Jensen (F), genitopatellar (G), Kabuki (H), Sotos (I), and Williams (J) syndromes.

Figure 5

Detection of Imprinting Defects and Trinucleotide Repeat Expansion Disorders using DNA Methylation Analysis The figure shows the DNA methylation pattern of four imprinted regions (A–D) and two coordinates with trinucleotide repeat elements (E and F) in 2,880 healthy individuals (blue) along with subjects identified with an imprinting or a trinucleotide repeat disorder (red and green). The x axis indicates the genomic coordinates and the y axis represents the DNA methylation levels between 0 and 1. Circles indicate the DNA methylation level for every individual at one CpG site. Neighboring CpG sites are connected with a line. (A) Two subjects with hypermethylation (red) and one subject with hypomethylation (green) in the promoter of SNRPN are shown, indicating a deviation from the normal imprinting pattern (hemi-methylation) in Prader-Willi and Angelman syndromes, respectively. (B) An extensive hypomethylation in the promoter of the imprinted gene KCNQ1OT1, indicating Beckwith-Wiedemann syndrome. (C) Hypomethylation in the imprinted region mapping to the promoter of MKRN3. (D) Hypomethylation of the promoter of MEG3 in two subjects. (E) Healthy males and females in the promoter of FMR1 show a hypo- and hemi-methylated pattern (blue), respectively. Three male subjects are shown with a hemi-methylation pattern (n = 1, mosaic event), and a full hypermethylated pattern (n = 2), an indication for the expansion of the CGG repeat beyond the normal count (fragile X syndrome). (F) Two individuals show increased methylation in the promoter of DIP2B, indicating CGG repeat expansion associated with mental retardation, FRA12A type. The same increased DNA methylation pattern is also observed in two apparently healthy subjects. Notably, the extended repeat in this region is not 100% penetrant.

Figure 6

Detection of Rare Epi-variants with Potential Explanation of the Unresolved ND/CA-Affected Case Subjects’ Phenotypes The figure shows the DNA methylation pattern of 4 selected rare epi-variants that are likely explanatory of the phenotypes observed in 6 subjects (red) compared with 2,880 healthy individuals (blue). The x axis represents the genomic coordinates and the y axis shows the DNA methylation levels between 0 and 1. Circles indicate the DNA methylation level for every individual at one CpG site. Neighboring CpG sites are connected with a line. (A) Hypomethylation in the promoter of PDE4D. (B) Hypermethylation in the tightly unmethylated promoter of STUB1. (C) Three individuals showing a hemi-methylation pattern in the normally hypermethylated promoter of CDC186. (D) Hypermethylation in the promoter of ELFN1.

Figure 7

Hypermethylation of the VRK2 Promoter Secondary to a 30 kb Upstream Deletion The figure shows a hemi-methylation pattern in a tightly unmethylated CpG island in the promoter of VRK2, secondary to a microdeletion ∼30 kb upstream. The deleted element is shown to harbor a site enriched for Histone 3 lysine 27 acetylation (H3K27ac), a marker of active regulatory elements. Affected case subjects, red; 2,880 healthy individuals, blue; x axis, genomic coordinate; y axis, DNA methylation levels between 0 and 1; circles, DNA methylation level for every individual at one CpG site.