NEW YORK – A year after announcing its intent to start selling next-gen sequencing systems in the US, China's MGI said last week that it is now ready to roll out the first instruments, which will use a new, antibody-based sequencing chemistry called CoolMPS that MGI maintains does not violate Illumina's patents.
Illumina has not taken kindly to those plans. Last week, the company asked a court in the US to issue a preliminary injunction to prevent MGI, a unit of BGI Group, from distributing its instruments to US labs on a no-cost trial basis.
At the Advances in Genome Biology and Technology meeting in Marco Island, Florida this week, MGI representatives provided additional details about the planned commercial rollout, and customers from Europe and Canada who already have an MGI instrument reported on their initial experience with the platform.
Although MGI has not presented a detailed side-by-side comparison of data from the CoolMPS chemistry with Illumina data, the company claims that it can not only reduce the cost of sequencing but also offer better data quality.
Last week, MGI researchers published a detailed description of the CoolMPS chemistry, which uses unlabeled reversible terminator nucleotides and four fluorescently labeled monoclonal antibodies to detect the bases, in a preprint in BioRxiv. During a conference presentation at AGBT, MGI CSO and BGI Senior VP Rade Drmanac explained that the antibodies are highly specific to one base, even though some of the bases differ only by a few atoms. Antibody binding is dependent on the terminator group, and once the terminator is removed, no molecular "scar" remains on the DNA because the bases are unmodified. Also, each antibody can be labeled with many dye molecules, which increases the signal, so smaller DNA nanoballs (DNBs) could be used in the future.
Drmanac also reiterated that the DNA nanoballs on the platform's arrays are created PCR-free using rolling circle replication instead, which avoids clonal errors and led to a 55 percent reduction in whole-genome sequencing indel errors in their study, compared to Illumina sequencing based on PCR clusters. In addition, the amount of "index hopping" or barcode switching that occurs on Illumina flow cells is greatly reduced, he said.
In their preprint, the researchers generated 400-base single-end reads and 2x150-base paired-end reads using the CoolMPS chemistry, and they produced data for DNBs with fewer than 50 template copies. In addition, they showed proof of principle for a sequencing scheme that uses only two instead of four different fluorophores. Two labeled antibodies are added at a time and four images are taken in total for this approach, called "4-color sequencing on 2-color imagers." Drmanac said that this results in an exceptionally low error rate, in particular for G and C bases, a feature he said could "drive the future detection of somatic mutations" in cancer.
In the US, MGI plans to start making the DNBSeq-G400RS, which is similar to the MGISeq 2000 model sold in China, available in April. That system, which can run two flow cells and has a data output of up to 700 Gb per run, has a US list price of $360,000. According to MGI, it can sequence a human genome for $700 in consumables costs, and Drmanac said the company has seen significant interest from US customers already.
In the third quarter, MGI plans to roll out the DNBSeq-T7 in the US, which it announced at AGBT last year and launched last fall. That system, which can run up to four flow cells and has a data output of up to 6 Tb per run with a one-day run time, has a US list price of $1 million. It promises to bring down consumables cost to $500 per human genome, even if run with just one flow cell to sequence 10 to 15 genomes. Since last September, MGI has installed 20 T7 systems in China and Asia Pacific, Drmanac said, and it plans to ship the first T7 to Europe soon.
The super-high-throughput DNBSeq-Tx, which MGI unveiled at AGBT this week, needs to be custom-ordered and it is not clear when the first system might arrive in the US. It can sequence up to 100,000 human genomes per year and promises to reduce the consumables cost per human genome to $100 when run at scale. A list price for the Tx has not been determined yet, so it is difficult to calculate the total sequencing cost per genome.
It is also unclear whether the DNASeq-T7 will reduce the true cost of sequencing a human genome by much, or at all. According to Shawn Levy, CSO of HudsonAlpha Discovery, who runs a high-throughput sequencing lab, Illumina NovaSeq customers who receive the largest reagent discounts can already get a human genome for $525 in consumables costs.
Factoring in amortization of the T7 over three-years, service contracts, and typical instrument failures would likely bring the "real world price" of a T7 human genome to somewhere around $600, he said, which would not include library prep or compute costs. "This would be a mild discount to Illumina but one [Illumina] could easily respond to," Levy said. "The key will be if [MGI] can show a read length and accuracy improvement via the scarless chemistry. That would be more impactful in the overall equation than just price."
In the meantime, several customers in Europe and Canada who already have an MGI instrument in their labs have started to generate data, and some presented results at AGBT this week.
MGI has installed platforms in several European countries so far, including at the Karolinska Institute in Sweden, the Institute of Cancer Research in the UK, VIB in Belgium, and the University of Tübingen and Saarland University in Germany. At least to some labs, it has made the DNBSeq-G400 available under a reagent-rental model in Europe, where customers do not pay for the instrument upfront but commit to purchasing reagents.
While MGI's manufacturing site in Latvia opened last November, it is still ramping up and the sequencers are currently manufactured at three factories in China. The Latvia site also serves as a demo lab for potential customers, though it does not perform service work.
Olaf Rieß' lab at the University of Tübingen has had the MGISeq-2000 (now the DNBSeq-G400) for a year and has used it for whole-genome sequencing of human genomes in various research projects. "The data quality of the PCR-free libraries is very high, on par with Illumina's NovaSeq, but without the index hopping," said Stephan Ossowski, a researcher in the Tübingen lab. Also, hands-on time for library prep is fairly similar for both technologies, he added.
He said his team has benchmarked the system with Genome in a Bottle standards and has seen "very little difference in the detected SNV and indel sets."
In addition, he and his colleagues have been able to readily use computational tools on the MGI data that were originally developed for Illumina data, which helped with the benchmark tests.
Recently, they have also combined short reads from MGI with long nanopore reads and optical maps from Bionano Genomics to generate high-quality human genome assemblies for less than $2,000. Ossowski said they are hoping to use such de novo assemblies in diagnostics to detect complex structural variants and to phase haplotypes.
Next, they intend to implement high-coverage exome sequencing for cancer genomics on the MGI sequencer. They also plan to acquire additional MGI instruments, "including the T7 to facilitate large-scale genome projects in Germany and Europe," he said.
Nik Matthews, a researcher at the Tumor Profiling Unit at the Institute of Cancer Research in London, reported during an MGI-sponsored workshop at AGBT on initial results his lab has obtained with the new CoolMPS chemistry. He said the lab was the first in the UK to obtain an MGI sequencer, the DNBSeq-G400, and has also tested the MGISP-100 automation system. Comparing the traditional MGI chemistry with the CoolMPS chemistry on the same PCR-free whole-genome sequencing MGI libraries, his team found that the latter generated higher-quality reads. The next project, he added, will be a bake-off with Illumina data.
Canada's Michael Smith Genome Sciences Centre at BC Cancer has had a DNBSeq-G400 since last October. Steven Jones, co-director and head of bioinformatics at the center, who presented during a second MGI workshop at AGBT, said his team has compared whole-genome sequencing data from the MGI instrument, using the standard chemistry, for a human cell line with Illumina HiSeq X data. They found that the MGI data scored "identical" or "slightly higher" than the Illumina data for several metrics. His lab has also been able to generate 150 x 175-base paired-end reads with the MGI, extending the second read, which has provided some extra sensitivity, accuracy, and better alignment.
The researchers have further sequenced tumor samples with both technologies and found that the MGI generated fewer duplicate reads than the Illumina and achieved comparable coverage. SNP and indel calls did overlap for variants with high allele frequencies but showed less overlap for rare variants. For RNA-seq, concordance between the two platforms was "as you would expect" and differences were "fairly negligible," he said. Also, structural variant detection and gene fusion detection has been similar between the two, and the lab has been "pleased with the ability to do single-cell sequencing," both transcriptome sequencing and ATAC-seq, he said.
Ioannis Ragoussis, head of genome sciences at the McGill University Genome Center, also talked about results from the DNBSeq-G400, which his lab received at the end of last year. So far, the lab has evaluated the platform for whole-genome sequencing, whole-genome bisulfite sequencing, and single-cell RNA-seq. It recently switched from an Illumina HiSeq 4000 to a NovaSeq and is interested in a sequencer that is more flexible and doesn't require so many libraries to be pooled per run in order to make it cost effective. For single-cell sequencing, he found that the metrics between the two systems were very similar, though the NovaSeq had a slightly higher percentage of valid barcodes than the DNBSeq. After clustering the cell types, "the end result between the platforms is really the same," he said. However, the MGI sequencer required less library pooling and provided more reads for the same cost.