3-hour genome sequencing and targeted analysis to rapidly assess genetic risk

Abstract

Purpose

Rapid genetic testing in the critical care setting may guide diagnostic evaluation, direct therapies, and help families and care providers make informed decisions about goals of care. We tested whether a simplified DNA extraction and library preparation process would enable us to perform ultrarapid assessment of genetic risk for a Mendelian condition, based on information from an affected sibling, using long-read genome sequencing and targeted analysis.

Methods

Following extraction of DNA from cord blood and rapid library preparation, genome sequencing was performed on an Oxford Nanopore PromethION. FASTQ files were generated from original sequencing data in near real-time and aligned to a reference genome. Variant calling and analysis were performed at timed intervals.

Results

We optimized the DNA extraction and library preparation methods to create sufficient library for sequencing from 500 μL of blood. Real-time, targeted analysis was performed to determine that the newborn was neither affected nor a heterozygote for variants underlying a Mendelian condition. Phasing of the target region and prior knowledge of the affected haplotypes supported our interpretation despite a low level of coverage at 3 hours of life.

Conclusion

This proof-of-concept experiment demonstrates how prior knowledge of haplotype structure or familial variants can be used to rapidly evaluate an individual at risk for a genetic disease. Although ultrarapid sequencing remains both complex and cost prohibitive, our method is more easily automated than prior approaches and uses smaller volumes of blood and thus may be more easily adopted for future studies of ultrarapid genome sequencing in the clinical setting.

Introduction

The benefits of rapid genetic testing in critically ill individuals have been demonstrated.1, 2, 3, 4, 5, 6 A precise genetic diagnosis helps guide management while giving families and providers valuable information to make informed decisions about goals of care.6, 7, 8 Because management decisions for critically ill individuals often must be made in hours or days, minimizing the time required to make a precise genetic diagnosis is of broad interest.

The turnaround time for rapid genetic testing via genome sequencing has decreased from approximately 26 hours in 2015 to just under 8 hours in early 2022.^1,4,9, 10, 11 This reduction in turnaround time has been enabled by the introduction of new sequencing technologies, advances in sequencing chemistry, and the development of analysis pipelines that can quickly and efficiently prioritize variants with limited manual input. Recently, Oxford Nanopore sequencing was used to rapidly evaluate a cohort of critically ill individuals with the shortest time to identification of a pathogenic variant in just under 8 hours.¹ Nanopore technology is an ideal platform on which to develop ultrarapid sequencing approaches because sequencing data from individual DNA molecules are available in near real-time.¹²

Here, we assess such an approach in a family in which prior clinical testing of a child with a suspected Mendelian disorder identified a single pathogenic variant in a gene associated with a familial condition (acrodermatitis enteropathica), but no second variant.¹³ Targeted long-read sequencing (LRS) performed on a research basis identified a candidate promoter variant on the opposite haplotype, which we confirmed affects gene expression in a cellular model. The family was interested in knowing whether a newborn biological sibling of the first child inherited the known pathogenic variants. Although treatment for this condition is not known to be beneficial immediately after birth,¹⁴ the case was chosen because it presented a proof-of-concept opportunity. Within 3 hours of birth, the genome of the at-risk neonate was evaluated using Nanopore sequencing followed by targeted analysis (Figure 1).

Figure 1.

Timeline from birth to result. The proband in this report was born at 22:37 PDT. Cord blood was collected in the delivery room and walked to the laboratory. DNA isolation and QC required 39 total minutes. Library preparation and loading of 10 flow cells was completed in 37 minutes. Once the first 10 flow cells were confirmed to be running well, 10 additional flow cells were loaded. At approximately 01:37 PDT, or 3 hours after the child was born, Clair3 was run and reads mapping to the SLC39A4 region were isolated and phased into 2 distinct haplotypes. This demonstrated that the newborn did not inherit either the known pathogenic or the candidate promoter variants.

We set forth to further decrease the time required to identify known pathogenic variants via genome sequencing by optimizing the DNA extraction and library preparation steps followed by targeted analysis based on prior genetic information. We also sought to use smaller blood volumes—an important consideration in neonates for whom there is an upper limit on the amount blood that may be drawn per day.¹⁵ Finally, although the process described here was performed in a research lab under ideal conditions, clinical adoption of ultrarapid sequencing will likely require automation of some steps. We therefore developed a protocol that required fewer manual touchpoints, which should be easier to automate.

Materials and Methods

Targeted LRS and analysis of sibling #1

Sibling #1 is the brother of the newborn and is affected with acrodermatitis enteropathica. DNA for sequencing was extracted from blood using the Monarch High Molecular Weight DNA Extraction Kit for Cells and Blood (NEB #T3050) following the suggested protocol for DNA isolation from blood with the following specifications: 500 μL of blood was used as input, shaking occurred at 900 rpm, bead binding time was extended to 8 minutes, and a sterile glass plating bead was used to homogenize the elution. DNA was sheared to an average fragment length of 10 kb using a Covaris gTUBE as described previously.¹⁶ Libraries for sequencing were prepared using the Oxford Nanopore Ligation Kit (SQK-LSK110) and loaded onto an R9.4.1 flow cell on a GridION. MinKNOW version 21.10.8 running Guppy 5.0.17 was configured to run adaptive sampling with a target region of approximately 2.6 Mb for approximately 48 hours (Supplemental Table 1). FASTQ files were generated with Guppy 5.0.12 (Oxford Nanopore) using the super accurate model (dna_r9.4.1_450bps_sup.cfg). Single-nucleotide and indel variants were called with Clair3,¹⁷ phased with LongPhase,¹⁸ and annotated with VEP 103.1¹⁹ using annotations from SpliceAI²⁰ and CADD v1.6.²¹

Plasmid construction for validation of the promoter variant in sibling #1

The pGL3-Basic-IRES firefly luciferase plasmid was a gift from Joshua Mendell (Addgene plasmid # 64784; RRID:Addgene_64784).²² The human SLC39A4 (HGNC:17129) proximal promoter/exon 1 (NC_000008.11:g.14416053-144417111) (Figure 2A) was amplified from genomic template isolated from the B-lymphocyte cell line GM04738 (Coriell) with forward and reverse (FW/RV) primers 5’ GCGCTAGCGCAAGAGCAAAACTCCAACTCG 3’ / 5’ GGCCTCGAGAGGCTCGCCCAGGCCCAG ‘3 and subcloned into the NheI/XhoI sites upstream of the IRES within pGL3-Basic-IRES. Site-directed mutagenesis was performed using the Quikchange-II site-directed mutagenesis kit (Agilent #200523) to generate the pGL3:SLC39A4/T (WT) construct by converting a common minor allele variant (rs117568226) within the cloned promoter (A) to the major allele (G) for consistency with the patient sample, which is homozygous for the major allele at this position (FW primer sequence 5’ GATTGGCTGCTGGAGATGCCTGGGTTAACCATTCC 3’). The pGL3:SLC39A4/T was then used as a template for site-directed mutagenesis to generate the pGL3:SLC39A4/C patient mutation construct (FW primer sequence 5’ GCAGAGCTCTGTGACTGGCTGCTGGAGATGC 3’). The pRL:TK renilla luciferase plasmid is commercially available (Promega #E2241). All plasmids were validated by Sanger sequencing (Genewiz).

Figure 2.

Establishing that the SLC39A2 promoter variant disrupts promoter activity. A. The promoter variant is located within a region that is evolutionarily conserved and demonstrates chromatin accessibility and transcription factor occupancy selectively within gastrointestinal tissues. The sequence variant compared with the reference sequence is shown. B. Luciferase assay of the SLC39A2 promoter region and variant promoter region in HEK293 cells demonstrating that the SLC39A2 promoter variant disrupts the promoter activity of this region, resulting in 21% promoter activity compared with the wild-type sequence.

We then transformed the pGL3:SLC39A4/T and pGL3:SLC39A4/C plasmids into NEB 5-alpha Competent E. coli (NEB #C2987H), which were grown from frozen glycerol stocks in LB broth + 100 μg/mL ampicillin. The pRL:TK renilla plasmid was transformed into NEB 5-alpha Competent E. coli, grown on LB agar + 100 μg/mL ampicillin plates, then a single colony was picked and grown in LB broth + 100 μg/mL ampicillin. All plasmids were purified with Qiagen Plasmid Maxi Kit (Qiagen #12162). Three separate maxi preps (each) were prepared for pGL3:SLC39A4/T and pGL3:SLC39A4/C.

Lipofectamine transfection for validation of the promoter variant

HEK293T cells (RRID:CVCL_0063) were a gift from Lea Starita’s lab and grown in Dulbecco’s modified Eagle’s medium (DMEM) (Thermo Fisher Scientific #11995-065) supplemented with 10% fetal bovine serum (HyClone #SH30396.03IH25-40) and 1% Penicillin-Streptomycin (Gibco #15140122) at 37 °C and 5% CO2 in T-75 flasks. Cells were split 1:5-10 at confluency, approximately every 2 to 4 days. The cells were seeded at ∼80% confluency in a 12-well plate (1 mL complete DMEM/well) and simultaneously transfected with luciferase vectors. Transient transfection was performed with Lipofectamine 3000 transfection reagent (Thermo Fisher Scientific #L3000015) per manufacturer’s protocol. 800 ng of experimental plasmid (pGL3:SLC39A4/T or pGL3:SLC39A4/C) or empty vector (pGL3-Basic-IRES) and 200 ng of pRL:TK renilla vector were transfected into each experimental well (in technical triplicate). Untransfected cells were seeded in triplicate. The experiment was replicated for 3 separate maxi preps of the experimental vectors and on 3 separate passages of HEK293T cells, for a total of 9 experimental replicates.

Dual-Glo luciferase assay for validation of the promoter variant

48 hours after transfection, media in the 12-well plate was replaced with 75 μL room-temperature complete DMEM per well, and luciferase assay was performed with the Dual-Glo Luciferase Assay System (Promega #E2920). Briefly, 75 μL of room-temperature Dual-Glo Luciferase Reagent was added to each well and agitated for 10 minutes at room temperature. The contents of each well were transferred to a white-bottom 96-well plate (Greiner Bio-One #655083) to measure firefly luminescence using a Synergy H1 plate reader (Biotek) with a 1 second integration time. Following this reading, 75 μL of Dual-Glo Stop and Glo Reagent was added to each well and incubated for 10 minutes at room temperature. The plate was then read for renilla luminescence with the same settings.

The average reading of the untransfected wells was subtracted from each experimental reading for firefly and renilla luminescence, respectively (Figure 2B). For each background-subtracted technical replicate, the firefly luciferase value was divided by the renilla luciferase value to normalize for transfection efficiency and viable cell number in each well, and this ratio was averaged between 3 technical replicates to calculate the normalized luciferase expression per plasmid. To calculate relative luciferase units, normalized luciferase expression from each experimental plasmid was normalized to the normalized luciferase expression from pGL3-Basic-IRES from the same cell passage, and the y-axis was scaled by dividing all values by the overall median of the normalized wild type. One experimental replicate was excluded from analysis because firefly luciferase values were at background levels. Statistical significance was analyzed by two-tailed Student’s t test.

DNA extraction and quantification from the newborn

At birth, 1 mL of cord blood was collected in an EDTA tube and placed on ice. DNA was extracted using the Monarch Genomic DNA Purification Kit (NEB #T3010) following the recommended protocol for genomic DNA purification from mammalian whole blood. Briefly, 5 individual reactions were prepared: 100 μL of whole blood was added to a 2-mL microfuge tube along with 10 μL of Proteinase K, 3 μL of RNase A, and 100 μL of Blood Lysis Buffer. This was vortexed and incubated for 5 min at 56 °C in a thermal mixer at 1400 rpm. 400 μL of gDNA binding buffer was then added to each tube and pulse-vortexed for 10 seconds. The combined lysate and binding buffer were transferred to a gDNA purification column in a collection tube. The tube was centrifuged at 1000g for 1 minute then 12,000g for 3 minutes. Columns were transferred to a new collection tube, and 500 μl of gDNA wash buffer was added, the cap was closed, and the collection tube was inverted 3 times. Tubes were then centrifuged for 1 minute at 12,000g, the flow-through was discarded, and an additional 500 μL of gDNA wash buffer was added and centrifuged again for 1 minute at 12,000g. The gDNA purification columns were transferred to 1.5-mL low-bind tubes and 61 μL of gDNA elution buffer preheated to 60 °C was added and allowed to sit for 1 minute at room temperature. The columns were then centrifuged for 1 minute at 12,000g. 1 μL of each extraction was quantified with the Qubit dsDNA HS (High Sensitivity) Assay Kit (ThermoFisher) (Supplemental Table 2).

Library preparation

Libraries for sequencing were prepared using a combination of reagents from the Oxford Nanopore Rapid Sequencing (SQK-RAD004) and polymerase chain reaction (PCR)-cDNA Sequencing (SQK-PCS111) kits in order to increase per-flow cell output. Four separate library preparation reactions were prepared in 1.5-mL Eppendorf tubes by combining 80 μL of genomic DNA with a target of 2–4 μg of DNA per reaction (Supplemental Table 3). 30 μL of nuclease-free water was added to each tube to bring the total volume to 110 μL. 12.5 μL of FRA (fragmentation buffer) was added to each tube and incubated at 30 °C for 2 minutes followed by 80 °C for 2 minutes, then placed in a cold block for 15 seconds. 5 μL of RAP-T (rapid adapter T) from the SQK-PCS111 kit was added to each tube and incubated for 5 minutes at room temperature. The RAP-T sequencing adapter from the SQK-PCS111 kit was used instead of the RAP (rapid) adapter from the SQK-RAD004 kit to take advantage of its higher pore occupancy rate, which results in higher output. After 5 minutes, 72.5 μL of nuclease-free water was added to each tube and the tube was placed on ice. The library mix for loading was made by adding 32 μL of library to a 1.5-mL Eppendorf lo-bind tube followed by 75 μL of SQB (sequencing buffer) and 43 μL of nuclease-free water, for a total volume of 150 μL.

Flow cell preparation, loading, and sequencing

During the DNA extraction and library preparation steps, 20 Oxford Nanopore R9.4.1 PromethION flow cells were prepared for sequencing. Flow cells were allowed to sit at room temperature for at least 10 minutes before being placed into a PromethION 24 running MinKNOW control software v22.03.4. Ten flow cells were primed with 500 μL of FB (flush buffer) + FLT (flush tether), then allowed to sit for at least 5 minutes before an additional 500 μL of FB + FLT was added to each. 150 μL of the library mix was then loaded to each flow cell and allowed to sit for 5 minutes before beginning sequencing. After confirming that the first 10 flow cells were working as expected, a second group of 10 flow cells were primed and loaded as described. Each experiment was configured to run using the protocol specified in the SQK-RAD004 kit, with reserved pores and live base calling turned off (Supplemental Table 4).

Data transfer and base calling

Original sequencing data were transferred to a remote base-calling server using rsync. Base calling was done using Guppy 6.2.1 (Oxford Nanopore) and the dna_r9.4.1_450bps_hac_prom.cfg model with a minimum quality score cutoff of 7 “--min_qscore 7.” A custom script was used to monitor for incoming sequencing data and perform sequential base calling of each library on a single NVIDIA A100 GPU. Base calling was performed on a machine with 4 NVIDIA A100 GPUs; thus, 4 libraries could be called simultaneously. After the first round of base calling for each library, the “—resume” flag was used to resume base calling from the last position for each library.

Alignment and analysis

A custom script monitored each library for new FASTQ files generated by the base calling process. New FASTQ files were combined into a single FASTQ file, then aligned to GRCh38 using minimap2.²³ After alignment, SAMtools²⁴ was used to extract reads surrounding the target gene, SLC39A4 (NC_000008.11:g.14400000-14450000 [chr8:14,400,000-14,450,000]) and 2 control genes; COL1A1 (NC_000017.11:g.50000000-50300000 [chr17:50,000,000-50,300,000]) and F8 (NC_000023.11:g.154800000-15200000 [chrX:154,800,000-155,200,000]). Individual BAM files for each target gene were merged with previous BAM files using SAMtools merge. Beginning at 3 hours after birth and every 30 minutes thereafter, variants were called on the combined bam file with Clair3¹⁷ followed by phasing using LongPhase.¹⁸ Variants within the region of interest were filtered from the vcf file and the phased bam file was visualized using Integrative Genomics Viewer (IGV).²⁵

Sequencing was stopped 7 hours after birth, or 5.5 hours after sequencing began, and all flow cells were washed and stored. Average coverage of chromosome 8 at 1-hour time points was monitored during sequencing (Supplemental Table 5). Variant calling statistics, phasing statistics, and switch errors were calculated using WhatsHap (Supplemental Table 6; Martin M, Patterson M, Garg S, et al. WhatsHap: fast and accurate read-based phasing. Biorxiv. Published online. 2016:085050. https://doi.org/10.1101/085050). The deeper coverage obtained by 7 hours of life allowed us to determine whether single-nucleotide variants (SNVs) identified at 3 hours of life and their haplotype assignments correlated with the later findings. At 7 hours of life, we identified a single SNV on HP:D that was not called at 3 hours of life (NC_000008.11:g.144413427 [chr8:144,413,427]) and no variants that had been assigned to a haplotype at 3 hours of life were assigned to a different haplotype at 7 hours of life. We did not attempt to analyze indel variants.

Results

The newborn and his genetic sibling (#1) were conceived using donated anonymized embryos from the same biologic parents. Sibling #1 developed a rash accompanied by increasing irritability of unknown etiology at 2 months of age that was initially attributed to eczema but failed to respond to standard treatment. Review of records revealed that 2 other genetic siblings (#2 and #3) had allergies and eczema, 1 of whom (#2) was also reported to have clinical zinc deficiency. Accordingly, sibling #1 was suspected to have acrodermatitis enteropathica (MIM: 201100), an autosomal recessive condition due to zinc transporter defect encoded by SLC39A4 (HGNC:17129), which results in reduced intestinal absorption of zinc. Consistent with this diagnosis, plasma zinc levels obtained at 8 months of age were low (0.26 μg/mL; normal range: 0.6-1.2 μg/mL). Sibling #1 was started on 3 mg/kg elemental zinc supplementation with near complete resolution of rash, irritability, and fatigue, and his zinc levels have been normal since beginning treatment. Sibling #1 received early intervention services because of late prematurity and torticollis, and at 22 months of age, has a persistent delay in expressive language (6-9-month level) but typical motor and social development.

Clinical testing of sibling #1 identified only a single pathogenic variant (NC_000008.11:g.144414042C>T [NM_130849.4:c.1203G>A, p.(Trp401∗)]) in SLC39A4. To identify a second pathogenic variant putatively missed by clinical testing, we performed targeted LRS of blood-derived DNA from sibling #1 and found a promoter variant (NC_000008.11:g.144416958T>C) on the other allele in an evolutionarily conserved CCAAT box that demonstrates chromatin accessibility and transcription factor occupancy selectively within gastrointestinal cells (Figures 2A and 3A).^12,16,26,27 The promoter variant disrupts the A at the 4th position of the CCAAT box and is predicted to be disruptive based on several predictive algorithms (ie, FINSURF score of 0.8936, and LINSIGHT score of 0.927754).^28,29 CCAAT boxes are essential components of many promoters and play a critical role in promoting transcription.³⁰ Thus, it was suspected that this variant represented a “second hit” in this individual through loss of transcription factor binding to the CCAAT box and subsequent loss of SLC39A4 transcription from this allele. Subsequent clinical exome sequencing and targeted variant testing in sibling #1 validated the presence of the promoter variant, with both tests reporting it as a variant of uncertain significance.

Figure 3.

IGV view of SLC39A4 variants found in the affected brother (sibling #1) (A), and the newborn at 3 hours (B) and 7 hours of life (C). Haplotypes denoted on the left are as listed in Supplemental Table 7. Indels of <3 nt have been hidden. A. The known pathogenic stop variant is present on HP:A (blue arrow), whereas the promoter variant is found on HP:B (yellow arrow). B. Phased reads from the newborn after approximately 1.5 hours of sequencing, or at 3 hours of life. Neither the previously known pathogenic single-nucleotide variant (blue box) or the promoter variant (yellow box) were observed in the newborn. C. After 5.5 hours of sequencing, or at 7 hours of life, neither of the previously known variants were identified. In both sibling #1 (A) and the newborn (B and C), nearly all reads span the 3-kb distance between the 2 variants.

To validate the impact of this variant, we performed a luciferase assay in HEK293 cells, which demonstrated that the promoter variant resulted in 21% promoter activity compared with wild type (Figure 2B), confirming the deleterious nature of the variant. Neither of the other 2 genetic siblings (#2 and #3) with eczema, allergies, and zinc deficiency were available for testing. Classification of this noncoding variant using ACMG criteria can be challenging given that biological parents or other affected siblings were not available for testing, as well as limited classification criteria for noncoding variation within the existing ACMG framework.³¹ Before functional testing, we applied PM2, PM3, PP3, and PP4, which gave a classification of likely pathogenic. It is important to note that, although biological parents were not available for testing, we were able to phase the variants using our LRS data and apply PM3. This is information that would not be available to a clinical lab and a likely explanation of why the variant was classified as a variant of uncertain significance by clinical testing. After functional testing we applied PS3 in addition to the above categories, changing the classification to pathogenic.

Before the second pathogenic variant in sibling #1 was known, the family elected to proceed with implantation of a randomly selected embryo from the same in vitro fertilization cycle as sibling #1. Although early treatment with zinc is not known to be beneficial for this condition, the family was interested in pursuing rapid genetic testing after birth to determine if the newborn had inherited either or both of the stop or promoter variants in SLC39A4 found in sibling #1. The neonate was born at 39-2/7 weeks by Cesarean section after a pregnancy complicated by polyhydramnios, large for gestational age, and decreased fetal movements in the third trimester. Prenatal testing included prenatal screening using cell-free DNA, which did not suggest an increased risk of aneuploidy. Birthweight was 3.526 kg (85th percentile), and Apgar scores were 6 at 1 minute and 8 at 5 minutes. The neonatal course was unremarkable, and he was discharged home at 2 days of age.

Cord blood was collected at birth, and ultrarapid long-read genome sequencing was performed in a research lab using 20 Nanopore PromethION flow cells with sequencing libraries prepared by a combination of 2 commercially available Nanopore kits in order to increase affinity of the library for the pore and thus increase flow cell output. The advantage of the rapid library protocol is that sequencing libraries can be generated in as little as 10 minutes using a transposase to insert adapter attachment sites followed by chemical attachment of a sequencing adapter. A possible limitation of this approach is that the random integration of the transposase into DNA can result in shorter overall DNA fragments than a ligation library preparation protocol, but the impact of shorter read lengths on outcomes when doing ultrarapid sequencing have yet to be evaluated. Accordingly, the average read lengths for our 20 sequencing runs ranged from 6 to 11 kb with batch effects from our combined library preparation reactions (Supplemental Table 4).

After 1.5 hours of sequencing (3 hours after birth), we evaluated the aligned data and found that, despite low coverage (11×), the 8-kb region that included SLC39A4 was completely phased and included 12 heterozygous polymorphisms (Supplemental Table 7). Among these polymorphisms, 6 were not found in sibling #1, and none were the pathogenic stop variant or the promoter variant found in sibling #1 by LRS (Figure 2B). We allowed sequencing to continue for an additional 4 hours (5.5 total hours of sequencing), generating approximately 45× coverage by 7 hours of life (Supplemental Table 5). Subsequent variant calling with Clair3 and phasing with LongPhase confirmed the findings from 3 hours of life (Figure 2C). Approximately 83% of heterozygous SNVs on chromosomes 1 to 22 identified after 1.5 hours of sequencing were present at 5.5 hours of sequencing, with 99.5% of those having been assigned to the correct haplotype (Supplemental Table 6). Subsequent targeted clinical testing of the newborn confirmed that he did not inherit the pathogenic stop variant. Targeted clinical testing for the promoter variant was not available.

Discussion

Here, we present genome sequencing and targeted analysis of an individual at risk of an inherited Mendelian condition. Although performed under artificial conditions in a research laboratory, our proof-of-concept experiment provides a framework for the development of larger studies to evaluate the utility of ultrarapid sequencing using automated protocols and smaller sample volumes. Although inexpensive methods exist to perform rapid genotyping using approaches such as multiplex real-time polymerase chain reaction, they are limited by the number of variants that can be targeted per experiment and require prior knowledge of the target variant.^32,33 The availability of untargeted ultrarapid testing in the critical care setting could lead to reduced health care costs by reducing the number of diagnostic evaluations and shortening the time required to definitive treatment.

Adoption of ultrarapid LRS in the clinical environment faces several hurdles. As performed, our protocol is likely cost prohibitive in many settings. As of mid-2023 the cost of a single PromethION flow cell is $600 to $900 (USD), meaning that sequencing using 20 flow cells would cost $12,000 to $18,000 (USD), along with approximately $1000 in other laboratory expenses per experiment. This cost was mitigated in our laboratory because the flow cells were washed and reused for other experiments, but this may not be possible in the clinical laboratory. Future improvements in flow cell output will lessen the cost, but it is unclear at what price point this approach will be feasible. Our results also benefited from being conducted in a controlled environment using staff with extensive experience. It may not be realistic to rely on the ability of laboratory staff to quickly and repeatedly perform the required extraction and library preparation steps then load tens of flow cells to generate sufficient data for rapid analysis. Instead, broader adoption will likely require automation of one or more of the DNA extraction, library preparation, and flow cell loading steps. Performing rapid data analysis can also be limiting for many laboratories because the available computational resources may not be sufficient to complete the alignment and variant calling steps quickly, especially when performed genome wide. Thus, improvements to variant calling algorithms will likely be required to replicate our results when doing untargeted evaluation.

Our analysis was simplified by several factors: (1) a focus on a single gene, (2) knowledge of the pathogenic and candidate variants in that gene, and (3) knowledge of neighboring SNVs defining the affected haplotypes in sibling #1. This is not unlike other clinical scenarios when a newborn is known to be at risk of inheriting familial variants that have caused disease in other family members. Often, sequence data are available for affected individuals and could be used in the same way we have used them here to perform rapid assessment. The use of LRS enabled us to evaluate the inheritance of both haplotypes at low coverage—an approach that may allow for evaluation of risk in other cases which a precise molecular diagnosis has not been made but a specific locus is suspected. For those individuals in which a candidate variant, gene, or locus is unknown, real-time variant calling and phasing could be performed, with results compared with a database of curated variants.¹

For some critically ill newborns with clinical findings, such as hyperammonemia, severe hypoglycemia, lactic acidosis of unknown etiology, or seizures in the first hours of life, an ultrarapid precise genetic diagnosis could be pivotal to guiding treatment decisions. For other newborns, the benefit remains unclear because much of the first 6 to 12 hours of life involve stabilization and assessment of illness. Thus, genetic testing results may not be considered until the newborn is stabilized and the family has considered what role ultrarapid results may have in their decision-making process.³⁴ Nonetheless, the approach described here can be applied broadly to other patient populations of all ages for which suspicion of a genetic etiology is high, and a precise genetic diagnosis could guide treatment choices.

Data Availability

Data that support the findings of this study will be uploaded to ANViL under accession number phs003047.

Conflict of Interest

Joseph Guillory, Androo J. Markham, and Daniel R. Garalde are employees of Oxford Nanopore Technologies (ONT). Miranda P.G. Zalusky, Jonas A. Gustafson, and Cate R. Paschal have received travel support from ONT. Danny E. Miller is on a scientific advisory board at ONT and has received travel support from ONT to speak on their behalf. Danny E. Miller and Evan E. Eichler are engaged in a research agreement with ONT. Evan E. Eichler is a scientific advisory board (SAB) member of Variant Bio, Inc. Danny E. Miller holds stock options in MyOme.