Genome-wide de novo risk score implicates promoter variation in autism spectrum disorder

Abstract

Whole-genome sequencing (WGS) has facilitated the first genome-wide evaluations of the contribution of de novo noncoding mutations to complex disorders. Using WGS, we identified 255,106 de novo mutations among sample genomes from members of 1902 quartet families in which one child, but not a sibling or their parents, was affected by autism spectrum disorder (ASD). In contrast to coding mutations, no noncoding functional annotation category, analyzed in isolation, was significantly associated with ASD. Casting noncoding variation in the context of a de novo risk score across multiple annotation categories, however, did demonstrate association with mutations localized to promoter regions. We found that the strongest driver of this promoter signal emanates from evolutionarily conserved transcription factor binding sites distal to the transcription start site. These data suggest that de novo mutations in promoter regions, characterized by evolutionary and functional signatures, contribute to ASD.

Figure 1.. Category-wide association study on 1,902 ASD families.

A) De novo mutations were identified in 7,608 samples from 1,902 quartet families, each including an ASD case and an unaffected sibling control. The mean genome-wide mutation rate, corrected for paternal age, is shown for cases and controls. B) Each mutation was annotated against 70 annotation terms in five groups, combinations of which defined 55,143 annotation categories (Table S3, Fig. S5). C) A category-wide association study (CWAS) shows the degree to which de novo protein-truncating variants (PTVs) in each category (points) are enriched in cases (right x-axis) or controls (left x-axis) against the statistical evidence for this enrichment (y-axis). Red lines show the threshold for nominal significance (p=0.05) and significance after correction for 6,711 effective tests (19). The red ‘X’ shows the category of all PTVs without other annotations. The equivalent CWAS is shown for: D) de novo missense; and E) de novo noncoding variants. Statistical tests: B-D) Binomial exact test, two-tailed.

Figure 2.. Enrichment of conserved promoters in cases.

A) After excluding categories with PTVs, which are known to have a strong contribution to ASD, a de novo risk score was generated using Lasso regression to distinguish cases and controls in the first 519 families and tested on 1,383 new families. The same risk score was tested considering 163 noncoding categories only and, based on the enrichment of promoter categories in the risk score, for 45 promoter categories and 118 noncoding categories without promoters (Table S5). B) Considering 1,855 promoter annotation categories with ≥7 mutations, 118 reached nominal significance, 112 of which had an excess of mutations in cases. C) The observation of 112 nominally significant case-enriched categories (red line) and 6 control-enriched categories (blue line) in B is compared to permuted expectation (grey distribution). Statistical tests: A) Lasso regression with permutation testing. B) Binomial, two-sided. C) Permutation testing.

Figure 3.. Mapping ASD association within promoter regions by annotation terms.

A) DAWN uses p-value correlations between 1,310 promoter categories with ≥20 mutations to define 47 clusters (nodes, with size representing the number of categories in the cluster). Evidence for ASD association is evaluated in the context of the local p-value correlation network (edges) to estimate false discovery rate (FDR). Enrichment is shown by color for the nine clusters with FDR≤0.01 (Table 1). B) The number of de novo mutations shared between these nine clusters and the annotation terms enriched in these clusters, is shown as a correlation with hierarchical clustering. The black boxes show the first five divisions based on hierarchical clustering with two large groups: Active Transcription Start Site (TSS) and Conserved Loci. The numbers of de novo mutations in each group are shown in parentheses. C) The size and relationship of the groups of promoter mutations identified in ‘A’ and ‘B’, based on de novo mutation counts. The number of mutations in each group is shown in parentheses. D) Estimates of relative risk based on the number of de novo mutations in cases and controls within each group. E) Considering mutations at Conserved Loci, the degree of enrichment of mutations in cases vs. controls (red line) is shown in relation to permuted expectation (grey distributions). The mean number of mutations per child is shown in parentheses on the left. Nominally significant uncorrected p-values are shown in red. F) Distribution of nonverbal IQ in cases with mutations at Active TSS (blue) and Conserved Loci (purple) promoters vs. cases with neither (grey). Cases with de novo PTVs were excluded from all groups. Statistical tests: A) DAWN. E) Permutation testing. F) Wilcoxon signed rank, two-sided. Boxplot in E and F shows the median (black line), interquartile range (white box), and a further 1.5 times the interquartile range (whiskers). DD: developmental delay; MF: Midfetal; REP: Roadmap Epigenome; TSS: Transcription Start Site; UTR: Untranslated Region.

Figure 4.. Relationship of conserved promoter mutations to the Transcription Start Site.

A) Frequency of Conserved Loci promoter mutations in cases and controls across the promoter region. B) Frequency of Conserved Loci promoter mutations in cases and controls at JASPAR transcription factor binding sites (TFBS) across the promoter region. C) Enrichment of Conserved Loci promoter mutations in cases, shown as relative risk, in sliding windows of 200bp across the promoter region. The purple line is the generalized additive model fit for relative risk and the 95% confidence interval is in grey. Ticks under the plot show individual mutations in cases (red) and controls (blue). D) The plot in ‘C’ is repeated for Conserved Loci promoter mutations at JASPAR TFBS. Statistical tests: A, B) Binomial, two-sided. Error bars show the 95% confidence interval (95%CI). TFBS: transcription factor binding sites; TSS: transcription start site.