Multiple recurrent de novo CNVs, including duplications of the 7q11.23 Williams syndrome region, are strongly associated with autism

Abstract

We have undertaken a genome-wide analysis of rare copy-number variation (CNV) in 1124 autism spectrum disorder (ASD) families, each comprised of a single proband, unaffected parents, and, in most kindreds, an unaffected sibling. We find significant association of ASD with de novo duplications of 7q11.23, where the reciprocal deletion causes Williams-Beuren syndrome, characterized by a highly social personality. We identify rare recurrent de novo CNVs at five additional regions, including 16p13.2 (encompassing genes USP7 and C16orf72) and Cadherin 13, and implement a rigorous approach to evaluating the statistical significance of these observations. Overall, large de novo CNVs, particularly those encompassing multiple genes, confer substantial risks (OR = 5.6; CI = 2.6-12.0, p = 2.4 × 10(-7)). We estimate there are 130-234 ASD-related CNV regions in the human genome and present compelling evidence, based on cumulative data, for association of rare de novo events at 7q11.23, 15q11.2-13.1, 16p11.2, and Neurexin 1.

Figure 1. Flow chart of CNV detection and confirmation in the Simons Simplex Collection (SSC)

CNV detection was optimized by qPCR analysis of 115 predictions (Table S1, Figure S1). Quality control was performed to check for identity error and data quality (supplementary methods). De novo detection was optimized by qPCR analysis of 403 predictions (Figure S1) leading to the threshold of ≥20 probes and refinement of the prediction algorithm. All de novo CNVs reported in the study were confirmed using qPCR with absolute quantification.

Figure 2. The burden of rare de novo CNVs and genes mapping within them in 872 probands and 872 matched siblings

A) % of individuals with ≥1 rare de novo CNV in probands vs. siblings. Red = deletions; blue = duplications for A to E. B) Total number of rare de novo CNVs in probands vs. siblings (2 probands and 1 sibling have more than one). C) Number of RefSeq genes (Pruitt et al., 2007) overlapping rare de novo CNVs in probands vs. siblings. D) % of individuals with ≥1 rare de novo CNV (as shown in A) split by sex. Specific comparisons and associated p-values are given. E) Number of RefSeq genes overlapping rare de novo CNVs (as shown in C) split by sex. F) The distribution of rare de novo CNVs by size in probands (green) and siblings (purple). The dashed vertical line represents the mean plus two standard deviations of the sibling events. G) The distribution of rare de novo CNVs by number of RefSeq genes. Statistical significance was calculated using: Fisher's Exact test (A, D), Sign test (B), Wilcoxon paired test (C), and Wilcoxon test (E).

Figure 3. Genotype-phenotype analyses of probands carrying rare de novo CNVs

A) The number of RefSeq genes within rare de novo CNVs (genes) vs. CNV size (size), with proband (red) vs. sibling (blue). The slope of the lines shows the fitted significant (p=1 × 10⁻⁹) relationship between genes and size and the difference between the lines shows the fitted difference for probands and siblings (p=0.025): on average probands have more genes within a rare de novo CNV for any given size. B) genes vs. size, with sex of subject encoded by color as noted (sex). The slope of the line shows the fitted significant (p=7 × 10⁻¹⁰) relationship between genes and size while the presence of only one line reflects the lack of significant difference by sex (p=0.20). C) ADOS Combined Severity Score (CSS), a measure of autism severity, against genes and by sex. The lack of a line indicates the absence of a significant relationship. D) Full-scale IQ (IQ) against genes and by sex. The slope shows that IQ declines as a function of genes in males (p=0.02, Wilcoxon test); there is no significant relationship in females. E) Boxplot with 95% confidence intervals for IQ by presence (yellow) or absence (blue) of a detected rare de novo event in the probands. F) Distribution of IQ in probands with (yellow, N=63) rare de novo CNVs and without (blue, N=1,061).

Figure 4. Confirmed recurrent rare de novo CNVs

A) All recurrent de novo CNVs identified in 1,124 probands and 872 siblings. The gene count is given when >6 RefSeq genes map to an interval; a complete listing of genes is presented in Table S4. The total number of de novo and matching inherited CNVs in probands and siblings are shown for deletions (Del) and duplications (Dup) in parentheses. B) LogR data for 4 de novo duplications and 1 control with no CNV (CT) in the 7q11.23 interval. RefSeq genes within this region are noted below the ideogram; the orange bars represent flanking segmental duplications. NCBI 36 (hg18) genomic coordinates are shown with the scale indicated. The LogR for all probes within the region is shown; LogR values greater than 0.15 are in blue (suggesting a duplication) while LogR values less than −0.15 are in red (suggesting a deletion). B allele frequency data is not shown but supports the presence of a corresponding CNV. The approximate boundaries of the CNVs are shown by the vertical dashed red line and blue arrows. C) LogR data for 6 duplications (4 de novo), 8 deletions (7 de novo), and 1 control with no CNV (CT) in the 16p11.2 interval. The ideogram and intensity plots are as in B. D) Overlapping rare de novo and rare inherited CNVs identified in the 16p13.2 interval. The brackets show the boundaries of RefSeq genes; 2 genes are in common between all 3 duplications: USP7 and C16orf72. The frequency of duplications in the DGV is shown in purple: the majority of the recurrent de novo region is not present in the DGV. E) Overlapping rare de novo and rare inherited CNVs identified in the 16q23.3 interval. A 34kb deletion overlaps a 5Mb deletion over a CDH13 exon (represented by ticks on the gene). The frequency of CNVs observed in the DGV is shown at the bottom in purple.

Figure 5. Burden of rare CNVs in 872 probands and 872 matched siblings

A) Bar graph showing the log (10) number of genes present in all rare CNVs binned by size (in Mb), with probands shown in green and siblings in purple. B) The data from A with confirmed de novo events excluded leaving only CNVs transmitted from a parent to offspring. C) Only confirmed de novo events are shown. D-F) The ratio (y-axis) of number of genes in probands vs. siblings for specific size thresholds (x-axis): D) all rare CNVs (transmitted and de novo); E) transmitted events; F) de novo events only. G-K) The total number of transmitted deletions (red) and duplications (blue) for probands and siblings for varying categories of CNV (shown above the graph). Definitions are described in supplementary methods. P-values (noted above the bars) are calculated using the Sign test and are not corrected for multiple comparisons. L-P) As in G-K, with number of RefSeq genes within the CNVs (y-axis). P-values (noted above the bars) are estimated using a two-tailed paired t-test and are not corrected for approximately 3,000 comparisons.

Figure 6. Pathway analysis of genes mapping within transmitted rare CNVs

A) The number of pathways with a corrected p-value ≤0.05 identified in probands (green) and siblings (purple) by the programs Metacore (GeneGo Networks) and DAVID (level 4 terms). The input consisted of 1,516 RefSeq genes found only in transmitted rare CNVs in probands and the 1,357 RefSeq genes found only in transmitted rare CNVs in siblings; p-values are from B and C. B) Permutation analysis to assess significance of the difference between probands and siblings. The 2,873 genes identified in probands or siblings were divided randomly between probands and siblings in the same initial proportions. The lists were submitted to GeneGo Networks and the difference between the number of pathways in probands and siblings was recorded. This process was performed 100 times and the image shows the frequency of the results. Only 4 events showed a difference ≥18 (the difference seen in A, vertical dashed line) yielding a p-value of 0.04. C) Permutation analysis to calculate the significance value with DAVID (level 4 terms) using the same methods as in B. A single result was ≥40 (the difference seen in A, vertical dashed line) giving a p-value of 0.01. D) All pathways with a corrected p-value ≤0.05 identified by GeneGo Networks for probands (green) and siblings (purple). The length of the bar represents the significance value on a logarithmic scale.

Figure 7. De novo and transmitted CNVs in 15q11.2-13

A 13Mb region is identified by the red box on the ideogram at the top. The Region overview identifies the RefSeq genes present within the interval and multiple segmental duplications (the colors identify regions of homology (Makoff and Flomen, 2007)). 5 of these segmental duplications are commonly referred to as BP1-BP5. Known CNVs identifies duplications (blue) and deletions (red) that have been reported in the literature; the alternating red and blue colors denote both deletions and duplications. Disease associations identifies regions with reported associations to four developmental and neuropsychiatric conditions (supplementary materials). Of note, BP2-BP3 deletions lead to Prader-Willi or Angelman syndrome. Transmitted and de novo CNVs shows the frequency of duplications (blue) and deletions (red) in the DGV and SSC populations. While CNVs overlying the segmental duplications are common, CNVs between the breakpoints are generally rare. De novo CNVs shows confirmed de novo CNVs in single individuals identified in this study and prior ASD studies (Itsara et al., 2010; Marshall et al., 2008; Pinto et al., 2010; Sebat et al., 2007). A) An enlargement of BP2-3 showing the relationship of de novo CNVs, genes, and common regions in the DGV. A small atypical duplication includes the genes MAGEL2, MKRN3, NDN (Itsara et al., 2010). B) An enlargement of BP4-BP5 showing similar data and methods as A. Removing the three Class 5A isodicentric chr15 events results in a non-significant p-value (p=0.62). C) An enlargement of the CHRNA7 region showing enrichment of duplications in probands (N=10) vs. siblings (N=3). The p-value is p=0.05 (Fisher's exact test); uncorrected for 3,667 comparisons; the rate of duplications in the DGV is similar to that seen in probands (Table S7).