De novo rates and selection of large copy number variation

Abstract

While copy number variation (CNV) is an active area of research, de novo mutation rates within human populations are not well characterized. By focusing on large (>100 kbp) events, we estimate the rate of de novo CNV formation in humans by analyzing 4394 transmissions from human pedigrees with and without neurocognitive disease. We show that a significant limitation in directly measuring genome-wide CNV mutation is accessing DNA derived from primary tissues as opposed to cell lines. We conservatively estimated the genome-wide CNV mutation rate using single nucleotide polymorphism (SNP) microarrays to analyze whole-blood derived DNA from asthmatic trios, a collection in which we observed no elevation in the prevalence of large CNVs. At a resolution of ∼30 kb, nine de novo CNVs were observed from 772 transmissions, corresponding to a mutation rate of μ = 1.2 × 10(-2) CNVs per genome per transmission (μ = 6.5 × 10(-3) for CNVs >500 kb). Combined with previous estimates of CNV prevalence and assuming a model of mutation-selection balance, we estimate significant purifying selection for large (>500 kb) events at the genome-wide level to be s = 0.16. Supporting this, we identify de novo CNVs in 717 multiplex autism pedigrees from the AGRE collection and observe a fourfold enrichment (P = 1.4 × 10(-3)) for de novo CNVs in cases of multiplex autism versus unaffected siblings, suggesting that many de novo CNV mutations contribute a subtle, but significant risk for autism. We observe no parental bias in the origin or transmission of CNVs among any of the cohorts studied.

Figure 1.

An example of de novo duplication (A) and deletion (B) detected by SNP array data. SNP array data from the father, mother, and child are displayed as shown in the pedigree in the lower-right portion of each panel. Each plot shows LogR Ratio (vertical bars), B-allele frequency (solid points), intrachromosomal segmental duplications in direct orientation (green blocks), inverted orientation (blue blocks), and interchromosomal segmental duplications (orange blocks). CNVs are highlighted by gray rectangles, contrasting the LogR ratio (red) and B-allele frequency (blue) with flanking regions (black). The de novo duplication (A) is characterized by an increase in the LogR ratio and altered clustering of heterozygote B-allele frequencies not seen in either parent. The de novo deletion (B) displays a decreased LogR ratio and loss of heterozygosity not observed in either parent.

Figure 2.

Observed frequency of de novo CNVs as a function of minimum CNV size.

Figure 3.

Comparison of de novo CNV size and potential underlying mechanisms. We classified de novo events into three categories: segmental duplication (SD)-mediated (CNV breakpoints were flanked by directly orientated SDs), SD at one breakpoint (a cluster of SDs one side), or no SDs (no SDs were identified). SDs were defined as segments >1 kbp and >90% sequence identity. (A) Using de novo events from the AGRE and asthma trios, the cumulative distribution (scatter plot) of CNV size and the frequency of CNV classes in each CNV size quartile are shown. Both SD-mediated and associated events are significantly enriched for de novo events above the median size (P = 3.1 × 10⁻⁷ and 0.018 respectively, two-tailed Fisher's exact test) where they account for 63% (24/38) of the events. (B) The analysis was repeated for a recent analysis of controls obtained from blood DNA (Stefansson et al. 2008). In this study, SD-mediated events were enriched for events above the median CNV size (P = 0.00964, two-tailed Fisher's exact test), while SD-associated events were not (P = 1, two-tailed Fisher's exact test).