SAN FRANCISCO (GenomeWeb) – In an effort to study B cell RNA transcript isoforms more completely, researchers at the University of California, Santa Cruz have developed a method that improves the accuracy of single-cell RNA sequencing on Oxford Nanopore's MinIon platform.
The team has filed for a patent on the method, which was published this month in the Proceedings of the National Academy of Sciences, and is especially interested in applying it to B cells, in particular to catalog B cell surface receptor isoforms, which are important immunotherapy targets.
Chris Vollmers, senior author of the study and an assistant professor of biomedical engineering at UCSC, said that he had previously used Illumina sequencing technology to study single cells. "B cells in particular are super finicky to work with, but every B cell is inherently unique in the transcriptome — they rearrange their genomes to make unique antibody transcripts," he said.
While Illumina sequencing worked to an extent to study these transcripts, the read lengths were too short to be able to study the cells at the transcript isoform level. "You can see what's expressed," Vollmers said, "but you can't look at isoforms, and most genes have multiple isoforms."
Thus, the team decided to turn toward sequencing technology with longer read lengths, and settled on the MinIon, primarily because of instrument cost. However, nanopore sequencing has two downsides compared to short-read technologies, like Illumina — lower throughput and lower accuracy. Also, most of the existing single-cell RNA-seq methods have been developed to work with short-read technology, so they rely on being able to generate highly accurate reads, Vollmers said.
"We wanted to co-opt all this development in the single-cell space, but apply it to nanopore sequencing," he said, but in order to do that, the group first needed to boost the read accuracy.
As described in the study, the researchers developed a method, which they dubbed rolling circle to concatemeric consensus (R2C2), that is similar to the circular consensus sequencing strategy employed by Pacific Biosciences. The strategy relies on circularizing cDNA molecules and then using rolling circle amplification to generate multiple copies of that molecule that are ligated together and can be sequenced in one read on the MinIon.
The technique was able to boost the read accuracy from around 87 percent to 94 or 95 percent, Vollmers said.
The researchers isolated single cells using FACS sorting and then used a modified version of the Smart-seq2 protocol, dubbed Tn5Prime, that the UCSC group described earlier this year in Nucleic Acids Research. The Tn5Prime method enhances coverage at the 5' end of transcripts.
After preparing RNA from single cells and generating barcoded cDNA molecules, the researchers then implemented the R2C2 method. First, they used a New England Biolabs' Gibson assembly method to prepare the cDNA for circularization. Gibson assembly is normally used for making synthetic DNA, but Vollmers said it also works well for circularization. The key to the Gibson assembly step, he said, is that only full-length cDNA molecules are circularized. That is also the main difference between R2C2 and the PacBio circular consensus sequencing method, Vollmers said, which uses standard ligation and so some of the circularized cDNA molecules are not full-length transcripts. Following circularization, the team amplified the DNA circles using rolling circle amplification.
"That generates a long piece of cDNA that contains multiple copies of your initial cDNA molecule," Vollmers said, which can then be sequenced on the MinIon. The researchers demonstrated that they were able to generate molecules between 10 kilobases and 50 kilobases in size, which Vollmers said represented, on average, about five copies of a cDNA molecule in one long piece.
Those molecules are then sequenced on the MinIon. The team also had to design bioinformatics tools that would enable them to figure out the original cDNA sequence from one long read of multiple cDNA molecules, and to do consensus calling. Vollmers said that compared to standard sequencing on the MinIon, which had a median base accuracy of 87 percent, the method had a median accuracy of between 94 and 95 percent, "which makes a world of a difference when you analyze isoforms in transcripts."
The researchers also compared their method with PacBio's Iso-Seq method for RNA sequencing. For that, they analyzed synthetic spike-in RNA controls with known variants using the R2C2 and Iso-Seq protocols. Overall, they found that they were able to generate around the same number of transcripts with both methods. Comparing the methods' abilities to identify splice sites and isoforms, they found that the R2C2 method matched the known annotated splice sites with 91 percent accuracy, a significant improvement over standard nanopore sequencing's 80 percent accuracy, but not as high as PacBio's accuracy of 97 percent.
Overall, Vollmers said, the method was a good compromise between cost, throughput, and accuracy. While the accuracy of the individual reads was not as high as compared to the PacBio Iso-Seq method, they were able to generate more reads. Tweaking the protocol to increase the accuracy would result in a lower throughput, he said. He estimated that the R2C2 method would not add significant cost to the overall experiment, noting that it just required standard off-the-shelf reagents. The turnaround time for the library prep is about a day and a half.
Next, the researchers tested the method on 96 B cells. The R2C2 method generated just over 1 million full-length consensus reads, about 730,000 of which aligned to the human genome and could be assigned to a single B cell. Importantly, the team was able to identify isoforms. They detected isoforms in more than 92 percent of genes that were covered by more than 10 reads.
Vollmers said that the team plans to follow up on one finding in particular — the multiple isoforms found in several B cell receptors. That is especially important because certain B cell receptors, like CD19 and CD20, are targets for a type of immunotherapy, CAR T cell therapy.
"It turns out that some of the B cells didn't have the binding site that's currently used," Vollmers said. "Even in healthy B cells, the binding site was alternative."
Vollmers said that his group plans to follow up on this finding by studying many more B cells, including from healthy and cancer populations. "In this study, we just looked at around 100 cells. But to get to the bottom of this, we would need to look at tens of thousands of cells from several people, as well as cells from B cell cancers, both pre- and post-therapy," he said, adding that he plans to apply for funding for such a study.
In terms of commercialization, Vollmers said he is not currently in any active discussions with companies about commercializing the method, although he acknowledged that developing the method into a kit form would make it easier to use. In addition, his group did apply for a patent on the method. However, he said that his main priority now is to get funding for larger studies.