High molecular diagnostic yields and novel phenotypic expansions involving syndromic anorectal malformations

Abstract

Evidence suggests that genetic factors contribute to the development of anorectal malformations (ARMs). However, the etiology of the majority of ARMs cases remains unclear. Exome sequencing (ES) may be underutilized in the diagnostic workup of ARMs due to uncertainty regarding its diagnostic yield. In a clinical database of ~17,000 individuals referred for ES, we identified 130 individuals with syndromic ARMs. A definitive or probable diagnosis was made in 45 of these individuals for a diagnostic yield of 34.6% (45/130). The molecular diagnostic yield of individuals who initially met criteria for VACTERL association was lower than those who did not (26.8% vs 44.1%; p = 0.0437), suggesting that non-genetic factors may play an important role in this subset of syndromic ARM cases. Within this cohort, we identified two individuals who carried de novo pathogenic frameshift variants in ADNP, two individuals who were homozygous for pathogenic variants in BBS1, and single individuals who carried pathogenic or likely pathogenic variants in CREBBP, EP300, FANCC, KDM6A, SETD2, and SMARCA4. The association of these genes with ARMs was supported by previously published cases, and their similarity to known ARM genes as demonstrated using a machine learning algorithm. These data suggest that ES should be considered for all individuals with syndromic ARMs in whom a molecular diagnosis has not been made, and that ARMs represent a low penetrance phenotype associated with Helsmoortel-van der Aa syndrome, Bardet-Biedl syndrome 1, Rubinstein-Taybi syndromes 1 and 2, Fanconi anemia group C, Kabuki syndrome 2, SETD2-related disorders, and Coffin-Siris syndrome 4.

Fig. 1. Machine learning allows all RefSeq genes to be ranked based on their similarity to genes known to cause ARMs.

A Receiver operating characteristic (ROC) curves were generated in validation studies of our machine-learning scoring approach. In this figure, colored ROC curves were generated using data from a single knowledge source, and the black ROC curve represents an omnibus score generated using the average score of all knowledge sources. The positive area underneath each curve indicates that our scoring approach identified training set genes known to cause ARMs more efficiently than random chance (diagonal dashed line). B After validation, ARMs-specific pathogenicity scores were calculated for all RefSeq genes. Box plots were generated based on the ARM-specific pathogenicity scores of (1) training set genes, (2) genes for which there is sufficient evidence to support a phenotype expansion involving ARMs (Table 1), and (3) genes for which there is currently insufficient evidence to support a phenotype expansion involving ARMs (Table 2). The median pathogenicity scores of the genes listed in Table 1 (83.3%) and Table 2 (70.5%) are lower than median pathogenicity score of the training set (98%) but exceed the median for all RefSeq genes (50%) indicated by the dashed line. This indicates that each of these groups is enriched for genes that are similar to the known ARMs genes in the training set. Epi = epigenetic histone modifications data from the NIH Roadmap Epigenomics Mapping Consortium, Exp = the GeneAtlas expression distribution, GO = Gene Ontology, MGI = the Mouse Genome Database, PINA = the Protein Interaction Network Analysis platform, TF = transcription factor binding data from the NIH Roadmap Epigenomics Mapping Consortium.