This article has been updated with a comment from Edico Genome on pricing for the Dragen system.
NEW YORK (GenomeWeb) – Researchers at the Center for Pediatric Genomic Medicine at Children's Mercy Hospital in Kansas City have improved the sensitivity, turnaround time, and scalability of their rapid whole-genome sequencing test, STAT-seq, for critically ill patients with suspected genetic disease.
In a study published online in Genome Medicine last night, the researchers described how they increased the test's sensitivity for nucleotide variants from less than 96 percent to 99.5 percent, and reduced the time to provisional results from 50 hours to 26 hours.
"The 26-hour genome is a proof of concept to see what could be done to push the boundaries of the technology," Neill Miller, director of informatics and software development at Children's Mercy and one of the lead authors of the study, told GenomeWeb.
Improvements included a modified Illumina HiSeq 2500 sequencer with faster sequencing cycles and Dragen data analysis technology from Edico Genome, which uses a hybrid hardware-software platform.
The study was led by Stephen Kingsmore, who recently left Children's Mercy to become the first president and CEO of the Rady Pediatric Genomic and Systems Medicine Institute, which is part of Rady Children's Hospital in San Diego.
What is most significant about the updated STAT-seq is not so much the reduction in turnaround time but the improved scalability and sensitivity, Kingsmore told GenomeWeb. "It's a much better test," he said.
Kingsmore and his Children's Mercy team published proof-of-concept for the original STAT-seq test three years ago. At the time, the turnaround time was around 50 hours, the estimated cost was $13,500, and the test had between 77 and 96 percent sensitivity as well as 99.5 percent specificity for the detection of nucleotide variants.
For the updated 26-hour test, the researchers sped up both the sequence data generation and the analysis component. Sample prep time — including DNA isolation, quality control, and shearing; PCR-free library prep; and library quality control — remained constant, at about seven hours.
Previously, cluster generation and paired-end sequencing on the Illumina HiSeq 2500 in rapid run mode took 25.5 hours. The researchers reduced that time to 18 to 21 hours, while maintaining sequence quality and quantity, by introducing faster sequencing cycles, fine-tuning the ramping of heating and cooling, optimizing temperature uniformity across the flow cell, and adjusting the microfluidics.
However, the bulk of the time savings came on the analysis side. With the first generation STAT-seq, sequence alignment, variant detection, and genotyping took about 15 hours, using gapped alignment and variant calling with Ilumina's Casava software. By switching to the Edico Genome Dragen pipeline, including the Dragen aligner and variant caller, the scientists were able to reduce that time to 40 minutes. This was possible, they wrote, because Dragen performs highly parallel alignments to a sorted reference genome and uses customized high-memory computer hardware.
The Children’s Mercy team also accelerated variant annotation for likely functional consequences, using the Rapid Understanding of Nucleotide variant Effect Software (RUNES), from 2.5 hours to 30 minutes, through "software refactoring," they wrote.
Finally, instead of manual analysis and interpretation of variants, the new test uses the interpretation software Variant Integration and Knowledge INterpretation In Genomes (VIKING), which allows for dynamic filtering of variants based on parameters such as clinical features, diseases, genes, pathogenicity citatory, allele frequency, genotype, and inheritance patterns.
Turnaround time for the test could be further reduced in the future. Customized robotics, they wrote, could bring the sample prep time down from the current 7.5 hours or so to two hours. Sequencing could come down from the current 18 to 20 hours to 10 hours if read lengths were shortened from 100 bases to 50 bases, while maintaining high specificity and sensitivity.
Having a test that takes less than 24 hours, they wrote, could make a difference because medical rounds "typically occur once a day," so test results returned by the morning could be discussed by the whole medical team. "That's the next barrier — same-day results," Kingsmore said.
The team tested the analytic performance of their 26-hour test protocols at two locations — Illumina's UK site and the genome center at Children's Mercy — over a period of two years, using several samples. Their fastest time to a provisional diagnosis, from the receipt of a blood sample, was 26 hours.
Children's Mercy has two modified HiSeq 2500 instruments that are capable of running in ultra-rapid mode. However, "some of the modifications of the sequencing process are not yet practical for day-to-day use in the lab," Miller said, so he and his colleagues "are routinely doing 30-hour whole genome runs using the standard 26-hour rapid mode on the HiSeq 2500 with many of the informatics improvements described in the manuscript."
Kingsmore said that "real-world" turnaround times for STAT-seq are on the order of two or three days right now, mostly because the lab currently does not operate around the clock, "so if a sample arrives at 4:00 pm, it won't be processed until the next morning."
The researchers also attempted to estimate the cost of the test. According to the study, reagent costs per sample were about $6,500, instrument depreciation was about $714 per genome, labor costs $70, computational costs $100, and interpretation and reporting costs ranged from $70 to $700, for a total on the order of $7,500 to $8,200 per sample. Sequencing of parent-child trios is usually needed for disease diagnosis. "Thus, cost is a significant barrier to broad adoption," they wrote.
Costs could be lower if the test was performed on a HiSeq X, they said, which would have a longer turnaround time of 41 hours, or if it was replaced by rapid exome sequencing, which could be possible in 36 hours at a cost of $6,500 per three trios, or just over $2,000 per trio.
"We are actively working on ways to create a ‘rapid mode’ for exome sequencing — the ideal would be an application that combines the low cost of exome sequencing with the rapid turnaround time of the ultra-fast whole genome," Miller said. "Primarily, this is a challenge for the lab, since the informatics solutions developed for the 26-hour genome can be used for analysis [of exome data] without further improvement."
Technically, the 26-hour whole-genome sequencing test could enable hospitals to scale it to larger numbers of patients. For example, each HiSeq 2500 sequencing instrument could run approximately 350 samples per year, the researchers wrote.
In addition, the Dragen hardware and software have "specifications which are likely to make genome sequencing practicable in many hospital laboratories," they noted, since it reduces the need for cloud computing or a large local cluster. Also, the VIKING software "greatly alleviates the burden of genome analysis and interpretation and allows common inheritance modes to be rapidly examined."
Kingsmore said the hardware component of the Dragen platform costs on the order of $25,000. "For many institutions, up until this Dragen technology became available, they were looking at needing $2 [million] to $3 million of supercompute to be able to process genomes. And now, suddenly, we have a 40-minute solution that will fit in anybody's IT budget." According to Edico Genome, Dragen is available under a platform-as-a-service model with tiered, undisclosed pricing that is based on data throughput.
In their paper, the researchers also pointed out a number of limitations of rapid whole-genome sequencing as a diagnostic tool. For example, the sensitivity of short-read sequencing for certain types of variants — such as structural variations and triplet repeat expansions — is currently low, though Kingsmore said this will likely change in the future.
Also, there might be gaps in coverage, and there is a lack of knowledge of all pathogenic variants. The biggest limitation, however, they wrote, is the interpretation of variants of uncertain significance.
Clinical applications of the 26-hour test would be those "that have a relatively high likelihood of guiding acute medical decisions in clinical situations where a delay is likely to result in significant morbidity or mortality," they wrote.
Children's Mercy has already been using slower versions of STAT-seq in a research setting for acutely ill infants in the neonatal intensive care unit. Earlier this year, they published a study of 35 infants in Lancet Respiratory Medicine, in which STAT-seq delivered a diagnosis in 57 percent of cases, of which 65 percent were useful for clinical management.
Kingsmore said he plans to scale STAT-seq testing up significantly, both at Children's Mercy and at Rady Children's Hospital, under research studies funded through the National Institutes of Health's Newborn Sequencing In Genome Medicine and Public Health (NSIGHT) program.