Poison exon annotations improve the yield of clinically relevant variants in genomic diagnostic testing

Neurodevelopmental disorders (NDDs) often result from rare genetic variation, but genomic testing yield for NDDs remains below 50%, suggesting that clinically relevant variants may be missed by standard analyses. Here, we analyze “poison exons” (PEs), which are evolutionarily conserved alternative exons often absent from standard gene annotations. Variants that alter PE inclusion can lead to loss of function and may be highly penetrant contributors to disease. We curated published RNA sequencing data from developing mouse cortex to define 1937 conserved PE regions potentially relevant to NDDs, and we analyzed variants found by genome sequencing in multiple NDD cohorts. Across 2999 probands, we found 6 novel clinically relevant variants in PE regions. Five of these variants are in genes that are part of the sodium voltage-gated channel alpha subunit family (SCN1A, SCN2A, and SCN8A), which is associated with epilepsies. One variant is in SNRPB, associated with cerebrocostomandibular syndrome. These variants have moderate to high computational impact assessments, are absent from population variant databases, and in genes with gene-phenotype associations consistent with each probands reported features. With a very minimal increase in variant analysis burden (average of 0.77 variants per proband), annotation of PEs can improve diagnostic yield for NDDs and likely other congenital conditions.