Regional nonsense constraint offers biological and clinical insights into genetic disease.

Reliably predicting the molecular impact of premature termination codons (PTCs) is essential for the clinical interpretation of "loss-of-function" variants in human disease. Measures of selective constraint can identify genes and genomic regions which are intolerant to deleterious genetic variation. However, existing loss-of-function constraint metrics do not comprehensively account for nonsense-mediated mRNA decay (NMD), a quality control pathway which critically regulates PTCs. Here, we use sequencing data from 730,947 individuals to develop an NMD-informed regional nonsense constraint metric. We find 2764 genes with significant regional nonsense constraint, including 641 known autosomal dominant disease genes. Using sequencing data in 32,260 trios from three rare disease cohorts, we find that de novo nonsense and frameshift variants are 9.5-fold enriched and associated with up to 5.9-fold higher odds of diagnosis in constrained regions versus unconstrained regions. We use these data to identify 22 candidate disease genes with clusters of de novo variants in constrained regions. These findings enhance clinical variant interpretation, deliver mechanistic insights in human disease, and empower the discovery of novel disease genes.