A recent study comparing Illumina's MiSeq and Life Technologies' Ion Proton platforms for low-coverage whole-genome sequencing to determine aneuploidy found that both platforms reliably detected fetal aneuploidies and determined sex but had differences in turnaround time, throughput, cost, and library preparation.
The study, led by researchers from BGI and published last month in PLoS One, evaluated the MiSeq and Proton platforms for accuracy in determining fetal aneuploidy, cost, speed, throughput, and other metrics that affect performance such as library preparation.
The researchers found that while both platforms called fetal aneuploidy with 100 percent sensitivity and specificity and showed similar levels of GC bias, the Proton was approximately eight times faster, three times less expensive, and generated about twice the data as the MiSeq.
The faster run time of the Proton, which "enables us to finish the detection within [one to two] working days, may bring benefits to applications with strict [turnaround time] requirements such as prenatal diagnosis," the authors wrote.
However, the researchers found that the Proton had a higher duplication rate in its raw data, which led to fewer effective reads — a problem, they indicated, that could be improved upon in the library preparation steps.
The researchers sequenced 18 genomes from spontaneously aborted fetuses, performing whole-genome sequencing to .1x coverage on both the MiSeq and Proton for each sample. While they did not test the platforms for their ability to sequence cell-free DNA from maternal plasma, BGI's Jiang Hui, senior author of the study, indicated that the method could have applications in prenatal diagnosis, as well as for determining the cause of fetal spontaneous abortion, preimplantation genetic diagnosis or screening, and for detecting chromosomal abnormalities in single cells.
Additionally, they tested Illumina's MiSeq instrument as opposed to the HiSeq — which is currently being used by companies offering noninvasive prenatal testing for fetal aneuploidy — due to the MiSeq's faster turnaround time. In clinical practice, "most of these sequence-based diagnoses have a strict turn-around-time requirement, especially for prenatal diagnosis," they wrote. "However, it takes several days for current high-throughput platforms to finish a sequencing run; for example, HiSeq 2000 will take approximately 11 days to generate 600 Gbp."
Hui added that both the MiSeq and Proton can generate over 5 Gb of data in one day, which "can meet most requirements for clinical labs."
The group collected samples from the Institute of Reproduction and Stem Cell Engineering of Central South University in China, and performed both comparative genomic hybridization and fluorescence in situ hybridization on them to confirm the fetal genotype. Each of the 18 fetal genomes carried a different aneuploidy.
The researchers used 50 ng of starting DNA for both platforms and performed 10 cycles of PCR during library preparation for both platforms, in order to compare GC bias on both.
For whole-genome sequencing on the MiSeq, the researchers performed standard 150-base paired-end sequencing for each sample. They mapped reads to the reference and used only unique non-duplicated reads for aneuploidy analysis. On the Proton, the researchers used the Ion One Touch 2 with a 200-base kit for library preparation and did single-end sequencing. They mapped reads to the reference, removed PCR duplicates, and used the uniquely mapped reads for aneuploidy analysis.
To detect fetal aneuploidy, the researchers used the Z-score method — essentially a counting method by which aneuploidy is called when the number of reads mapping to the chromosome of interest exceeds a certain threshold. Z-scores above three are called as aneuploidy.
The team did a 150-base paired-end sequencing run on the MiSeq, which took 27 hours and generated 4.56 million reads, averaging 38.15 megabases of sequence per sample. After they removed PCR duplications, they had around 214,000 paired reads per sample, or about 84.24 percent of the total.
On the Proton, the researchers generated 39.33 million single-end reads with a median length of 111 bases. After removing PCR duplications, they had around 1.7 million unique reads per sample, or 78.3 percent of the total.
While the Proton resulted in about an eight-fold increase in data compared to the MiSeq, it also had a much higher duplication rate, around 10 percent, compared to .08 percent on the MiSeq.
To compare quality metrics of the reads, the researchers selected 90,000 raw reads from each sample. From this set, duplication rate was .62 percent on the Proton and .07 percent on the MiSeq.
Since the researchers were comparing the platforms' ability to call copy number variants, they decided to evaluate coverage evenness, since that is one of the main factors that impact the ability to call copy number variants. From the 90,000-read subset, they found no significant differences between the two platforms. Additionally, the platforms showed comparable levels of GC bias.
Both the MiSeq and Proton called each of the different aneuploidies in all 18 samples with 100 percent specificity and sensitivity, but had differences in throughput, speed, and cost.
Hui added that the study showed that both MiSeq and the Proton could be used for aneuploidy detection based on ultra-low coverage sequencing and that they each have their advantages. "Ion Proton has better performance on turn-around time and cost, but MiSeq got more effective data with a lower duplication rate, under the equivalent data set," he said in an email.
One Proton run generated around 1.7 million uniquely mapped reads per sample, while one MiSeq run generated around 214,000 reads per sample. According to Edwin Cuppen, a professor of genome biology at the Hubrecht Institute in the Netherlands who was not involved in the study, for Z-score analyses, the greater the number of unique mappable reads, the more accurate the results, since the reads act as "independent measurement points." As such, he said that he was surprised the authors did not focus more on the differences between the number of reads per sample generated by each platform. "Most previous studies indicated that up to 10 million reads per sample provide the most robust results," he said.
Cuppen's lab offers noninvasive prenatal testing of fetal trisomies 21, 18, and 13 for women at higher risk for fetal aneuploidy on Life Technologies' SOLiD platform with the Wildfire technology, which enables library preparation to be done directly on the flow cell free of emulsion PCR.
Additionally, he added, for fetal aneuploidy detection, read length is less important than number of informative reads. "It is true that longer reads are in principle more informative than shorter ones because they can be mapped more easily to unique locations in the genome, but this advantage rapidly decreases above 50 bp in read length," he told In Sequence in an email.
He added that while the study is interesting and highlights the "potential of both the MiSeq and Proton," he would also be interested in seeing data from samples with variable and low fetal fractions and whether that impacts either of the platforms' accuracy in calling aneuploidy. "The largest challenge is to accurately detect aneuploidy with low fetal DNA contributions," he added.
The Proton also had a faster turnaround time than the MiSeq. Run time for the Proton was three to four hours compared to 24 to 27 hours for the MiSeq. The shorter run time enabled a total turnaround time of one to two days, the authors wrote. However, they did not specify total turnaround time for the MiSeq.
James Hadfield, head of the genomics core facility at Cancer Research UK, said that one aspect not mentioned in the study was that template preparation for the Proton has to be done off the sequencing instrument and adds an additional eight hours to the run time, while MiSeq template prep is done on the instrument and is included in the 24-to-27-hour total run time. Nevertheless, the Proton would still complete a run faster than the MiSeq. Hadfield also wrote about the study on his CoreGenomics blog.
Mike Lelivelt, director of bioinformatics at Ion Torrent said this faster run time would apply when comparing the Proton not just to the MiSeq, but to all of Illumina's next-gen sequencing instruments. The HiSeq 2500 or even the recently launched NextSeq 500 have the "same time profile," he said. Illumina's NextSeq 500 can do a 2x75 bp run in around 15 hours, or a 2x150 bp run in around 26 hours, while the HiSeq 2500 can do a 1x36 bp run in seven hours, with turnaround time increasing to 40 hours for a 2x150 run. Run time for the MiSeq also decreases to around four hours when doing a 1x36 bp run.
Illumina declined to comment.
Finally, the authors found that cost per sample was significantly less on the Proton than the MiSeq. Between 14 and 16 samples could be pooled together on one $750 MiSeq run, for $46 to $53 per sample, the authors wrote. On the Proton, they ran 70 to 75 samples on one $1,000 run for $14 to $15 per sample.
Moving forward, it remains to be seen whether labs will adopt the Proton for low-coverage, whole-genome sequencing clinical applications like noninvasive prenatal testing, but according to the PLoS One authors, it could be done faster than current NIPT offerings and has the potential to be as accurate, although they tested the protocol on fetal tissue, not cell-free DNA from maternal plasma.
"Our study has demonstrated the capability of [an] ultra-low coverage sequencing strategy in clinical applications, [but] more samples are required to make a more comprehensive conclusion," the authors wrote.