Copy number variants (CNVs) analysis in a deeply phenotyped cohort of individuals with intellectual disability (ID)

Abstract

Background: DNA copy number variants (CNVs) are found in 15% of subjects with ID but their association with phenotypic abnormalities has been predominantly studied in smaller cohorts of subjects with detailed yet non-systematically categorized phenotypes, or larger cohorts (thousands of cases) with smaller number of generalized phenotypes.

Methods: We evaluated the association of de novo, familial and common CNVs detected in 78 ID subjects with phenotypic abnormalities classified using the Winter-Baraitser Dysmorphology Database (WBDD) (formerly the London Dysmorphology Database). Terminology for 34 primary (coarse) and 169 secondary (fine) phenotype features were used to categorize the abnormal phenotypes and determine the prevalence of each phenotype in patients grouped by the type of CNV they had.

Results: In our cohort more than 50% of cases had abnormalities in primary categories related to head (cranium, forehead, ears, eye globes, eye associated structures, nose) as well as hands and feet. The median number of primary and secondary abnormalities was 12 and 18 per subject, respectively, indicating that the cohort consisted of subjects with a high number of phenotypic abnormalities (median De Vries score for the cohort was 5). The prevalence of each phenotypic abnormality was comparable in patients with de novo or familial CNVs in comparison to those with only common CNVs, although a trend for increased frequency of cranial and forehead abnormalities was noted in subjects with rare de novo and familial CNVs. Two clusters of subjects were identified based on the prevalence of each fine phenotypic feature, with an average of 28.3 and 13.5 abnormal phenotypes/subject in the two clusters respectively (P < 0.05).

Conclusions: Our study is a rare example of using standardized, deep morphologic phenotype clustering with phenotype/CNV correlation in a cohort of subjects with ID. The composition of the cohort inevitably influences the phenotype/genotype association, and our studies show that the influence of the de novo CNVs on the phenotype is less obvious in cohorts consisting of subjects with a high number of phenotypic abnormalities. The outcome of phenotype/genotype analysis also depends on the choice of phenotypes assessed and standardized phenotyping is required to minimize variability.

Figure 1

Data processing workflow.

Figure 2

Prevalence of abnormal coarse phenotypes. Thirty-four coarse phenotypes were evaluated among our 78 patients based on WBDD criteria (see Additional file 1: Table S1 for the whole term of each phenotype). *indicates phenotype with >95% or <5% prevalence in the cohort which was removed in the statistical analysis.

Figure 3

Phenotype and de novo CNV association analysis. Prevalence of the abnormality of each of the coarse phenotypes in individuals with de novo CNVs (18 cases) compared to individuals with only common CNVs (40 cases). The phenotypes with a prevalence >95% or <5% in the whole cohort (78 cases) were excluded from calculation.

Figure 4

Phenotype and familial CNV association analysis. Prevalence of abnormal coarse phenotypes in individuals with familial CNVs (20 cases) compared with those containing only common CNVs (40 cases). Two individuals with both de novo and familial CNVs were removed from the analysis. The phenotypes with a prevalence >95% or <5% in the whole cohort (78 cases) were excluded from calculation.

Figure 5

Clustering of individuals based on 80 fine phenotypes. (A) Data displayed as heat map. K-means method was used to group the 78 individuals into two clusters. The filled dark squares indicate an abnormal phenotype. Statistically significant differences in the number of phenotype abnormalities were found between the two clusters (P < 0.05, Wilcoxon rank-sum test). The different groups of CNVs in each individual are indicated at the top of the heat map. (B) Data displayed as barplot. The prevalence of individuals with an abnormal phenotype was compared between the two clusters. *indicates P < 0.05 (Fisher exact test after multiple test corrections).