Introduction:
Developmental disorders exhibit high genetic heterogeneity, often with a constellation of non-specific syndromic features, posing major challenges for molecular diagnosis. While exome and genome sequencing efforts have improved diagnostic yield, more than half of individuals with rare disease and their families remain on diagnostic odysseys. The limitation of short-read technologies in assessing the full repertoire of disease variants, such as complex structural variants, likely contributes to the high negative result rate of genetic testing. Long-read genome sequencing (LR-GS) using high quality PacBio HiFi read technology holds promise for uncovering missed etiologies.

Methods:
To assess this, we recruited children (and their unaffected parents) with undiagnosed rare developmental disorders with suspected genetic cause and prior negative or inconclusive genetic testing. For all participants, genomic DNA extracted from fresh whole blood was sequenced using a PacBio Revio system to a target mean coverage of 30X. For analysis, we developed a robust pipeline that leverages both phased assembly-based and read-based tools to call a wide range of variant types including single nucleotide variants (SNVs), structural variants (SVs), and repeat expansions. Phenotype-driven variant prioritization using Human Phenotype Ontology (HPO) terms was performed using a mixture of tertiary analysis solutions, including publicly (SvAnna) and commercially (Illumina’s Emedgene Software) available tools. To rule out potential candidate SVs based on frequency in the general population, we relied on LR-GS population data from the Human Genome Structural Variation Consortium version 2 (HGSVC2), the Consortium of Long Read Sequencing (CoLoRS) Database, and the Genome Answers for Kids Database (GA4K).

Results:
In summary, 16 of 17 trios were analyzed with LR-GS and 1 of 17 was analyzed with short-read GS only due to limited DNA availability. We identified reportable findings for research-use-only (RUO) in 64.7% (11/17) of analyzed trios. Among cases with reported findings, 58.8% (10/17) had small variant findings and 5.9% (1/17) had structural variant findings. Of the total 18 reported RUO variants across the 17 cases, 6 variants (from 3 cases) are in the process of clinical confirmation, and 1 variant has been clinically confirmed, which was previously detected on a prior clinical short-read sequencing test but was not reported. Notable positive findings included a case with compound heterozygous variants in TRAK1 and strong phenotypic overlap with developmental and epileptic encephalopathy 68 (MIM:618201). Two cases had strong phenotypic matches with recently characterized disorders. One case was heterozygous for a de novo missense variant in PAK2 implicated in Knobloch syndrome 2, and the other case was heterozygous for a de novo nonsense variant in DDX17 implicated in a neurodevelopmental disorder. Lastly, two cases had variants in genes of uncertain significance: a de novo frameshift variant in PTPDC1 and a homozygous ~50 kb deletion spanning ZNF705D, FAM66D, USP17L2, and USP17L7.

Conclusion:
Use of a multi-pronged analysis approach was key to the incorporation of long-read genomic data for rare disease. We observed an abundance of prioritized small and structural variants that were inherited from unaffected parents, underscoring the need for trio data in rare disease analysis. Several prioritized SVs by our pipeline that were absent from short-read-based SV population datasets, such as gnomADv4, were excluded due to an observed allele frequency higher than expected for disease in LR-GS population databases, demonstrating the need for LR-GS population database filtering in clinical pipelines. Additionally, our findings highlight the necessity of continual clinical genome reanalysis for variant discovery and the difficulty of structural variant interpretation in rare disease.