The analysis of de novo events in affected individuals lends support to both of these mechanisms: the CNVs in females are indeed significantly larger (with median of 10 selleck screening library genes per CNV in females, three genes per CNV in males, two-tail Mann Whitney, p value = 0.02), and genes derived from female CNVs are more functionally important for the network shown in Figure 2. Using simulations of random CNVs we also confirmed that the difference in the relative importance
of female versus male nodes is unlikely (p = 0.024) to be a simple consequence of the larger CNV sizes in females (see Supplemental Information; Figure S2C). We believe that both of the aforementioned Selleckchem Doxorubicin mechanisms are at play. Indeed, it would be surprising that stronger perturbation can be inflicted exclusively by larger CNVs and not disruption of high impact genes, and vice versa. Analysis of the established annotation resources, such as Swiss-Prot (UniProt Consortium, 2007), GeneCards (www.genecards.org), WikiGenes (www.wikigenes.org), and IHOP (Hoffmann and Valencia, 2004), suggests that a significant fraction of genes in the identified network either play a well-defined
functional role in the brain or have been previously implicated in neurodegenerative and psychiatric disorders. Only ∼25% (54 of a randomly selected 214) of all genes within the de novo CNV regions have been previously associated with brain-related phenotypes. However, when we consider genes in the identified clusters this proportion rises drastically (p value < 10−3), to ∼67% (Figure 2A; 30 out of 45) for the one-gene-per-CNV cluster or ∼52% (Figure 2B; 38 out of 72) for the two-genes-per-CNV cluster (see Table S2 for functional description of cluster genes). To characterize in more detail the specific biological processes
related to the cluster in Figure 2A, we investigated the strength of functional interactions between the cluster genes and various gene ontology (GO) categories (Ashburner et al., 2000). GO categories represent a CTP synthase curated set of functionally related genes described by a controlled vocabulary. For human genes in each of 1454 GO categories we calculated their average log likelihood interaction score (using the background network) with the genes in the identified cluster (Figure 2). The GO-specific significance of these interaction scores was calculated by comparison with scores of randomly generated CNV events with the same gene count at in real data by Levy et al. (2011). A false discovery rate (FDR) procedure was used to correct for multiple hypothesis testing (see Experimental Procedures). The 25 GO categories with lowest Q values, indicating the highest connection significance to the autism associated cluster, are shown in Table 1 (see Table S3 for other significant GO categories).