Impact of low-frequency coding variants on human facial shape

The contribution of low-frequency variants to the genomic architecture of normal-range facial traits is unknown. Therefore, we studied the influence of 31347 low-frequency coding variants (MAF < 1%) in 8091 genes on multi-dimensional facial shape phenotypes in a European cohort of 2329 healthy individuals. Using three-dimensional facial images, we partitioned the full face into 31 hierarchically arranged segments to model global-to-local features, and generated multi-dimensional phenotypes representing the shape variation within each segment. We used MultiSKAT, a multivariate kernel regression approach to scan the exome for face-associated low-frequency variants in a gene-based manner. After accounting for multiple tests, seven genes (AR, CARS2, FTSJ1, HFE, LTB4R, TELO2, NECTIN1) were significantly associated with morphology of the cheek, chin, nose and philtrum. These genes displayed a wide range of phenotypic effects, with some impacting the full face and others affecting localized regions. Notably, NECTIN1 is an established craniofacial gene that underlies both syndromic and isolated forms of cleft lip and palate. The missense variant rs142863092 in NECTIN1 had a significant individual effect on chin morphology, and it is predicted bioinformatically to be deleterious on the nectin-1 protein. We show that the zebrafish nectin1a mutation affects craniofacial development and leads to abnormal size and shape of the palate and Meckel's cartilage. These results expand our understanding of the genetic basis of normal-range facial shape by highlighting the role of low-frequency coding variants in novel genes.