This article was originally published Nov. 11.
Researchers from the University of California, San Diego have devised a single-cell sequencing method that makes use of nanoliter-sized microwells to reduce the persistent problem of amplification bias in single-cell sequencing.
Compared to the most complete single-cell Escherichia coli genome published using a conventional multiple displacement amplification method, the UCSD approach was able to recover 50 percent more of the genome with three to 13-fold less sequencing data. Additionally, they estimated that generating a library with their method was around 70 percent less expensive than using conventional MDA.
The researchers demonstrated their technique in an online publication in Nature Biotechnology this week, showing that it could sequence an entire E. coli genome from a single cell and identify copy number variations in single human neurons.
They now have a pending patent application on the method, and senior author Kun Zhang, an associate professor of bioengineering at UCSD, told In Sequence that they are interested in partnering with a company to commercialize it.
The key to the technique, dubbed MIDAS, for microwell displacement amplification system, is that the MDA reactions occur in tiny, nanoliter-sized wells, Zhang told IS. The researchers designed microwell arrays the size of a standard microscope slide, with each slide containing 16 arrays of 255 microwells, 400 microns in diameter, which allows for parallel amplification of 16 separate cell populations.
To ensure that each well contained only one cell, the researchers tested a number of different cell-loading densities, settling on a loading density of one cell per 10 wells, which resulted in no more than 0.5 percent of the wells containing more than one cell. The remaining empty wells were then used as controls to detect contamination.
Once the cells were seeded in the wells, MDA was carried out and the resulting amplicons were extracted for sequencing.
Zhang said that while the exact reasons why reducing the volume in which MDA reactions take place results in less bias are not quite clear, the phenomenon was demonstrated as early as 2006 by Stephen Quake's group at Stanford University. One hypothesis, he said, is that bias is reduced because there are fewer reagents for the same amount of starting material. Some regions of the genome tend to amplify more quickly than others for various reasons, including primer binding efficiency, but with fewer reagents those early events "may not have as many resources to grow faster," he explained, which "leaves a chance that the later events will catch up" and amplification will be more even across the genome.
With the MIDAS technique, the MDA reactions occur in 12-nanoliter wells. By comparison, a similar approach developed by Quake's group and commercialized by Fluidigm as their C1 system, uses microfluidic wells 135 nanoliters in size, Zhang said.
After seeding the wells and performing MDA reactions in each of the wells, the amplicons are extracted for library preparation and sequencing. Zhang said that creating libraries from the amplicons was one of the more challenging steps.
Zhang said that following MDA, the resulting DNA molecule takes on a hairball or onion-like structure. The researchers were trying to use Illumina's Nextera library preparation for low-input DNA, but Zhang said that the method at first didn't work, possibly because the transposases could not get at the center of the onion.
To get around this problem, Jeff Gole, previously a PhD student in Zhang's lab, who now works at Good Start Genetics, developed an "onion peeling protocol," Zhang said. Essentially, the method denatures the DNA to generate a linear fragment and then through enzymatic reactions creates a double-stranded molecule "with no weird structures," Zhang said.
The resulting DNA is then subjected to library preparation using the Nextera method and 100-base paired end sequencing on Illumina's MiSeq.
The researchers sequenced three single E. coli cells using the protocol, generating 2 million to 8 million reads per cell, corresponding to a genome coverage of 87x to 364x.
The genomes of each cell were assembled using the SPAdes algorithm. For each cell, between 88 percent and 94 percent of the genome was assembled with N50 contigs ranging from 2,654 bases to 27,882 bases and maximum contig length between 18,465 bases and 132,037 bases. More than 80 percent of the reads were mapped to the E. coli genome, with the remaining reads mostly common MDA contaminants.
Compared to standard MDA, MIDAS demonstrated less amplification bias and was able to capture 50 percent more of the genome with less sequencing — approximately 90x to 120x coverage compared to 1,200x coverage.
The researchers also tested the method on human neuronal cells for its ability to call copy number variants. They used nuclei from post-mortem brain samples, generating six sequencing libraries from four neurons from a Down syndrome individual and two neurons from a disease-free individual.
After sequencing, they used a binning approach to call CNVs, dividing the sequencing reads into nearly 50,000 60-kilobase bins. They also tested 240-kilobase bins, which reduced amplification bias even further but resulted in lower-resolution analysis.
Looking at conventional MDA, the researchers found it was not possible to distinguish copy number variants from random amplification bias, but with MIDAS, they were able to call the copy number variation in the Down syndrome cells at a 1-megabase to 2-megabase resolution.
Other groups have also been looking to improve upon single-cell sequencing methods, including Sunney Xie from Harvard University, who recently developed a method called multiple annealing and looping-based amplification, or MALBAC, which has been commercialized by Yikon Genomics; Quake's microfluidic-based approach that has since been commercialized by Fluidigm; a group at Cold Spring Harbor Laboratories that has developed Cell-Seq; and an approach designed by researchers from the Karolinska Institute for single-cell RNA sequencing dubbed Smart-seq; among others (IS 1/2/2013, IS 5/15/2012, and IS 10/8/2013).
The UCSD group compared MIDAS to publications of Quake's microfluidic technique, MALBAC, and standard MDA and found that MIDAS "compared favorably to each amplification method, generating the lowest levels of bias across the genome," the authors wrote.
Zhang said that team is now working on further optimizing the protocol. For instance, he said the team would like to increase seeding efficiency to enable more usable microwells and to automate various steps.
"Right now, the bottleneck is that we have to use a micromanipulator to suck up each amplicon" after amplification. Automation of that process and of the library preparation steps would "make things more consistent and higher throughput."
Additionally, he said the team is interested in partnering with a company to commercially manufacture the arrays, which would help make the method more accessible to other researchers.