Spatial Clustering of de Novo Missense Mutations Identifies Candidate Neurodevelopmental Disorder-Associated Genes

Abstract

Haploinsufficiency (HI) is the best characterized mechanism through which dominant mutations exert their effect and cause disease. Non-haploinsufficiency (NHI) mechanisms, such as gain-of-function and dominant-negative mechanisms, are often characterized by the spatial clustering of mutations, thereby affecting only particular regions or base pairs of a gene. Variants leading to haploinsufficency might occasionally cluster as well, for example in critical domains, but such clustering is on the whole less pronounced with mutations often spread throughout the gene. Here we exploit this property and develop a method to specifically identify genes with significant spatial clustering patterns of de novo mutations in large cohorts. We apply our method to a dataset of 4,061 de novo missense mutations from published exome studies of trios with intellectual disability and developmental disorders (ID/DD) and successfully identify 15 genes with clustering mutations, including 12 genes for which mutations are known to cause neurodevelopmental disorders. For 11 out of these 12, NHI mutation mechanisms have been reported. Additionally, we identify three candidate ID/DD-associated genes of which two have an established role in neuronal processes. We further observe a higher intolerance to normal genetic variation of the identified genes compared to known genes for which mutations lead to HI. Finally, 3D modeling of these mutations on their protein structures shows that 81% of the observed mutations are unlikely to affect the overall structural integrity and that they therefore most likely act through a mechanism other than HI.

Figure 1

Examples of Identified Genes with Clustering Mutations Protein domains are annotated based on Pfam HMM search. cDNA locations of de novo missense mutations are depicted by blue pins. Genes shown here are as follows: SMAD4 (A), CDK13 (B), PACS2 (C). Figures visualizing the clustering of de novo missense mutations in the other 12 genes are provided in Figures S1–S15.

Figure 2

Intolerance to Missense Variation Violin plots show the distribution of the gene-based $d_N/d_S$ (y axis) per gene set (x axis). The median $d_N/d_S$ is indicated by a red horizontal line. The NHI genes are more intolerant to missense variation than HI genes (HI genes median: 0.460; NHI genes median: 0.428; p = 2.24e-03). In addition, the identified genes with clustering mutations are more intolerant to missense variation than HI genes (genes with clustering mutations median: 0.352; p = 8.45e-03).

Figure 3

Examples of Modeling of Missense Mutations on 3D Protein Structures Wild-type residues are marked in blue; de novo mutations are indicated as red globes or lines (Tables S17). (A) 3D structure of GNA1, acting through HI, showing that the modeled missense mutations are buried and likely to disrupt protein folding. (B) Structure of PPP2R5D, acting through NHI, where the modeled missense mutations affect mostly surface residues and are expected to have no or only local structural effects. (C) Zoom-in of known missense variants p.Arg496Cys and p.Ile500Val in SMAD4 known to act through a gain-of-function mechanism. These variants are located on the surface of the monomer and in contact with another SMAD4 monomer. (D) Zoom-in of the missense variant p.Gly343Arg in ACTL6B which is located at the surface. The side-chain points toward the solvent, therefore the larger Arginine will fit. (E) Zoom-in of the missense variant p.Pro65Leu in PCGF2 close to the interaction site with other molecules.