Long-read genome sequencing and variant reanalysis increase diagnostic yield in neurodevelopmental disorders

Abstract

Variant detection from long-read genome sequencing (lrGS) has proven to be more accurate and comprehensive than variant detection from short-read genome sequencing (srGS). However, the rate at which lrGS can increase molecular diagnostic yield for rare disease is not yet precisely characterized. We performed lrGS using Pacific Biosciences "HiFi" technology on 96 short-read-negative probands with rare diseases that were suspected to be genetic. We generated hg38-aligned variants and de novo phased genome assemblies, and subsequently annotated, filtered, and curated variants using clinical standards. New disease-relevant or potentially relevant genetic findings were identified in 16/96 (16.7%) probands, nine of which (8/96, ∼9.4%) harbored pathogenic or likely pathogenic variants. Nine probands (∼9.4%) had variants that were accurately called in both srGS and lrGS and represent changes to clinical interpretation, mostly from recently published gene-disease associations. Seven cases included variants that were only correctly interpreted in lrGS, including copy-number variants (CNVs), an inversion, a mobile element insertion, two low-complexity repeat expansions, and a 1 bp deletion. While evidence for each of these variants is, in retrospect, visible in srGS, they were either not called within srGS data, were represented by calls with incorrect sizes or structures, or failed quality control and filtration. Thus, while reanalysis of older srGS data clearly increases diagnostic yield, we find that lrGS allows for substantial additional yield (7/96, 7.3%) beyond srGS. We anticipate that as lrGS analysis improves, and as lrGS data sets grow allowing for better variant-frequency annotation, the additional lrGS-only rare disease yield will grow over time.

Figure 1.

A de novo, 4 Mb paracentric inversion in proband 1, affecting ZBTB20. (A,B) Visualization of a subset of proband and parent reads in IGV at the 5′ (A, Chr 3:110,477,108–110,477,357) and 3′ (B, Chr 3:114,639,120–114,639,284) breakpoints (black arrowheads) indicate a de novo event. Reads in A show the SV is present on the proband's paternal allele. (C) Schematic of the proband's inversion. Section B–C is inverted, and junctions 1 and 2 (jct1, jct2) are displayed. The inversion disrupts ZBTB20, with a promoter (P) and exons 1–6 of NM_001348800.3 (yellow boxes) remaining at the distal end of the chromosome near jct2/D and exons 7–12 of NM_001348800.3 (green boxes) moving upstream to the breakpoint at the more proximal end of the chromosome (near jct1/A). Sequence level resolution of the inversion is also shown. (D) Alignment of the proband's assembled paternal contig versus the reference genome supports the inversion.

Figure 2.

Two ALS2 deletions in trans in proband 2. (A) Visualization of proband and parent reads in IGV (Chr 2:201,719,400–201,741,000) indicate two overlapping deletions in ALS2; a smaller maternal deletion (pink bar) and a larger paternal deletion (blue bar/arrow). Note that the three breakpoints shown here overlap Alu sequence in the RepeatMasker track. Alignment of the proband's assembled maternal (B) or paternal contig (C) versus the reference genome supports the two deletions (red dashed lines).

Figure 3.

Proband 3 has a 4 kb insertion in the 3′ UTR of HCFC1. (A) The proband's family has a history of X-linked ID, as the proband (arrow) and two other male relatives (gray squares) are affected. (B) Model of the relative length of the insertion in the 3′ UTR NM_005334.3. Only exons 24–26 are shown. (C) The hemizygous insertion is likely inherited from a heterozygous carrier mother, as indicated by srGS reads. Note that some reads contain unaligned ends (multicolored bases) while some span the entire insertion (purple line at the 5′ end of the target site duplication [TSD]); all the proband's reads support the insertion. A TSD is a hallmark of an L1-mediated insertion.