Large-scale targeted sequencing identifies risk genes for neurodevelopmental disorders

Abstract

Most genes associated with neurodevelopmental disorders (NDDs) were identified with an excess of de novo mutations (DNMs) but the significance in case-control mutation burden analysis is unestablished. Here, we sequence 63 genes in 16,294 NDD cases and an additional 62 genes in 6,211 NDD cases. By combining these with published data, we assess a total of 125 genes in over 16,000 NDD cases and compare the mutation burden to nonpsychiatric controls from ExAC. We identify 48 genes (25 newly reported) showing significant burden of ultra-rare (MAF < 0.01%) gene-disruptive mutations (FDR 5%), six of which reach family-wise error rate (FWER) significance (p < 1.25E-06). Among these 125 targeted genes, we also reevaluate DNM excess in 17,426 NDD trios with 6,499 new autism trios. We identify 90 genes enriched for DNMs (FDR 5%; e.g., GABRG2 and UIMC1); of which, 61 reach FWER significance (p < 3.64E-07; e.g., CASZ1). In addition to doubling the number of patients for many NDD risk genes, we present phenotype-genotype correlations for seven risk genes (CTCF, HNRNPU, KCNQ3, ZBTB18, TCF12, SPEN, and LEO1) based on this large-scale targeted sequencing effort.

Fig. 1. Overview of study design.

Targeted sequencing was performed in probands for two gene panels: NDD1 (63 genes) and hcNDD (62 genes). Gene and variant counts are after QC. The same categories of variants were retrieved from three previously published smMIP studies for 62 hcNDD genes. All smMIP variants were combined; redundant samples were eliminated and compared to the same category of variants from ExAC non-psych controls. The number of variants is after the exclusion of false positive variants and variants with insufficient coverage in ExAC. Mutation burden analysis identified 48 FDR significant genes (q_mutBurden < 0.05, Benjamini–Hochberg correction for 125 genes), of which six reached FWER significance (p_mutBurden < 1.25E−06, Bonferroni correction for 20,000 genes and two tests); DNMs of the 125 genes used in this study were identified from exome sequencing in 10,927 published NDD trios and 6,499 new ASD trios that combined as 17,426 NDD parent–child trios. A separate de novo enrichment analysis, using two statistical methods (CH model and denovolyzeR), identified 90 FDR significant genes (q_dnEnrich < 0.05, Benjamini–Hochberg correction for 18,946 genes in CH model and 19,618 genes in denovolyzeR), of which, 61 genes reach FWER significance (p_dnEnrich < 3.64E−07, Bonferroni correction for 19,618 genes and seven tests) for excess DNM. There is a significant overlap (40 genes) of the significant genes suggested by the two approaches. Then we performed genotype–phenotype correlation analysis for seven NDD risk genes (CTCF, HNRNPU, KCNQ3, ZBTB18, TCF12, SPEN, and LEO1) and present a clearer clinical picture of each gene.

Fig. 2. Significant genes identified from mutation burden and de novo enrichment analyses.

a Mutation burden analysis identified 48 genes significant for LGD and/or MIS30 variants in smMIP sequencing compared with the ExAC (r0.3) non-psych subset controls; each dot indicates a gene and the color indicates the category of variant showing significance for the gene (red for LGD, blue for MIS30, and black for both LGD and MIS30). b The CH model and denovolyzeR show high concordance for genes with significant excess of DNM at both FDR and FWER levels. c A union set of 90 genes showing excess DNM (FDR 5%) in de novo enrichment analysis. Gray dashed box in top panel is shown in bottom panel for a zoom view. See Supplementary Data 10 for underlying data.

Fig. 3. Severe variants and the genotype–phenotype correlations in CTCF.

a LGD (red) and MIS30 (blue) variants are depicted against a protein model for CTCF. Variants new to this study are shown above the protein while published DNMs from denovo-db (v1.5) are below. Variants are flagged with yellow lightning bolt if de novo. Annotated protein domains are shown (colored blocks) for the largest protein isoforms. b Heatmap depicts the common clinical features for patients carrying CTCF severe variants by using the specific HPO annotation (rows), which were retrieved from published studies and our cohort (columns). Phenotypic enrichment is shown according to the features’ recurrence labeled by the increment of color degree. The items with no data available were labeled with “-” and were excluded in the frequency analysis.

Fig. 4. Distribution of severe patient variants in six genes.

Protein diagrams are shown for HNRNPU (a) KCNQ3 (b) ZBTB18 (c) TCF12 (d) SPEN (e), and LEO1 (f) with the same display metrics that applied in Fig. 3. Validated LGD (red) and MIS30 (blue) variants are plotted. Variants listed above the protein model are new to this study, while the ones below were published previously. Paternal (black arrow) and maternal (green arrow) inheritance are shown if determined. A yellow lightning bolt denotes a de novo mutation.