One surprising finding is that a large proportion of genes with variants that we initially believed would have high phenotypic impact failed to associate with any classic phenotypes, or even with molecular endophenotypes. Among 41 confirmed nonsense variants with high predicted impacts, 18 nonsense variants failed to associate with any endophenotypes (across scans of 16 transcriptome data sets in different tissues) or with classic phenotypes at q<0.01. However, complete inactivation of four of these genes—Scn5a, Aimp1, Peli3 and Dlgap5—is known to cause severe phenotypes (MGI database, www.informatics.jax.org). Inactivation of Scn5a reduces embryo size and is associated with abnormal cardiovascular system function23. Complete inactivation of Aimp1 produces delayed wound healing and decreased inflammatory response24. The other two genes with nonsense mutations, Peli3 and Dlgap5, are linked to decreased viral infection25 and female infertility26, respectively. Failure to detect associated phenotypes could be interpreted as false negative results or inadequate phenome coverage, but we suspect that most commonly this reflects molecular resilience that buffers the phenotype from apparently strong homozygous mutations. For example, a tandem duplication in the cardiac actin gene (Actc1) reduces
