A Stanford-led research group is continuing to use microfluidics as the foundation of its high-throughput single cell sequencing and analysis studies, citing the system's potential for increasing automation and curbing the non-specific amplification of DNA from these individual cells.
Microfluidics allows for highly parallel and automatic experiments. But more importantly from a single-cell analysis perspective, this system makes it possible to perform multiple displacement amplification of single-cell DNA in a very small reaction volume, increasing the performance of this amplification step, explained Jianbin Wang.
Wang is a graduate student in Stephen Quake's Stanford University bioengineering lab and co-first author on a study published in the latest issue of Cell. In that study, Wang, Quake, and colleagues describe how they used microfluidics to genotype and sequence the DNA of around 100 single sperm cells from the same 40-year-old Caucasian individual. These experiments helped create a personal recombination map for the man and to deduce his germline de novo mutation rate and instability levels.
The study is the latest in a string of microfluidics-based single-cell and single-chromosome studies by the Stanford team that have used technology marketed by Fluidigm, of which Quake is a co-founder. Wang said the team did not collaborate with Fluidigm to develop a commercial version of the chip used in the current study, but noted that the company is expected to introduce a very similar chip, known as the C1 Single-Cell AutoPrep System, relatively soon.
"The most important advantage of microfluidics is that it improves the amplification performance in terms of efficiency and specificity," Wang told In Sequence.
The use of microfluidics does not alter the performance of phi29 polymerase — an enzyme used in the MDA reaction that appears to be at least as accurate as versions of the Taq polymerase enzyme that have proofreading capability.
Wang argued that problems seen with MDA in other single-cell studies have stemmed from reaction volumes that are too large, leading to spurious, non-specific products. The use of a microfluidic device reduces the MDA reaction volume to 50 nanoliters — about 1,000 times smaller than the volume used for conventional MDA reactions — thereby decreasing the noise associated with this amplification step.
"If you assume that you only have one cell, the additional volume would not improve your yield," he said. "The additional volume would have to deal with non-template. And your reagent is not 100 percent clean: there is random contamination and random primer dimers can produce non-specific products."
"I think it's a great study and there is some really interesting biology here on aneuploidy in single sperm cells, recombination rates, and detecting gene conversion during meiosis," MD Anderson Cancer Center researcher Nicholas Navin, who was not involved in the study but has been pursuing single-cell sequencing strategies, told In Sequence in an e-mail message.
Nevertheless, Navin he is not convinced that decreasing the MDA reaction volume produces a corresponding drop in the error associated with the phi29 polymerase. To that end, he said, he'd like to see a direct comparison of sequencing data stemming from single-cell DNA amplified by MDA in large and small reaction volumes in parallel.
At the Biology of Genomes meeting earlier this year, Navin outlined the Cell-Seq method he and his colleagues are developing to reduce the amplification needed for individual cells by specifically targeting cells at a stage in the cell cycle in which DNA has doubled through replication (IS 5/15/2012).
Also speaking during the same Biology of Genomes session was Albert Einstein College of Medicine researcher Adam Auton, whose team is doing its own recombination studies using multiple single sperm cells isolated from the same individual.
For the current study, the Stanford researchers initially attempted to isolate and amplify DNA from 125 single sperm cells using the Fluidigm 48.48 Genotyping Dynamic Array, a 48-well microfluidic chip specifically designed for this study. After removing samples from empty wells, they did recombination mapping from genome-wide, array-based genotyping data generated for 91 of the sperm cells.
More than three dozen sperm cells, including some used in the recombination mapping analysis, were sequenced to a range of coverage levels to look at genome stability and de novo mutation profiles, respectively.
The same chip was used to isolate and amplify individual chromosomes as part of a Chinese hamster ovary cell genome sequencing study published in Nature Biotechnology last year, while another similar chip was employed in a single-cell haplotyping study published in the same journal in 2010 (IS 12/21/2010).
Quake, who currently chairs Fluidigm's scientific advisory board, has also been involved in sequencing studies on individual bacterial cells that relied on the same sorts of microfluidic approaches.
To ensure that they were dealing with individual sperm cells that had been channeled into each well of the chip, researchers examined them microscopically, Wang explained, a step that is generally sufficient for experiments involving larger, diploid single cells, which are clearly visible under the microscope.
For the small, haploid sperm cells, though, the team did another check to be certain that they were not inadvertently looking at more than one cell at once. After one round of amplification by MDA, DNA from each well was genotyped by TaqMan PCR at 46 carefully selected sites in the genome.
These 46 positions are known to be highly polymorphic in humans, Wang explained, and any person selected at random is typically heterozygous at between 10 and 20 of these sites. In sperm cells, though, sequences should be homozygous at every position in the amplified DNA, owing to the haploid nature of these cells.
"Because we are dealing with haploid sperm cells, every single position should be homozygous," he said. "If there are two cells in the well that was amplified, there will be some positions that appear to be heterozygous because they're a mix of the genotype from the two cells."
Of the 93 MDA-amplified sperm cells that were subsequently sent out for genome-wide genotyping on the Illumina Omni1S Bead Array, two were found to have whole-chromosome deletions and were excluded from the resulting recombination map to avoid biasing the results.
The recombination map that was created from genotype data on the remaining 91 sperm cells indicated that the individual tested had genome- and chromosome-wide recombination patterns at the genome and chromosome levels that resembled those gleaned from population data on individuals of European descent.
For instance, sites of recombination were over-represented around telomeric regions of chromosomes but scarce near their centromeres, as reported in the past.
The team's finer-scale recombination analyses uncovered some differences in the individual's recombination patterns as well, including recombination events outside of previously described hotspots and variability in hotspot usage from one cell to the next.
On the whole-genome sequencing side, researchers used the Illumina GAII to sequence 31 individually isolated and amplified sperm cells to low coverage — around 0.02X — to look for large-scale genomic changes and to assess genome stability in the cells. The sperm cells in that analysis included eight of the 93 that had been genotyped in the recombination analysis, along with almost two-dozen other single sperm cells.
In addition to the two cells with missing chromosomes that had been found during the genotyping stage of the study, the team tracked down six single sperm cells that seemed to show genomic abnormalities.
Eight more sperm cells were sequenced to higher coverage with the Illumina HiSeq 2000 to determine the individual's de novo mutation rate. In contrast to de novo mutation studies done with family trios, which generally involve both parents and one child, the sperm cell approach offers a peek at the mutation profiles within several cells from the same individual at a specific age, Wang explained.
The mutation level was quite similar across all eight sperm cells, ranging from 25 to 36 apparent point mutations per cell, the researchers reported, suggesting that de novo mutation rates are fairly well maintained within each person at a specific time.
Despite the improvements in MDA accuracy achieved using a reduced reaction volume, there were still some gaps in the whole-genome sequence data used for that analysis, owing to variable amplification efficiency at different sites in the genome.
For the eight sperm cells tested by deeper whole-genome sequencing, the researchers generated sequence data covering around 30 percent to 50 percent of the genome at a depth of around six to eight fold.
That level of coverage is far shallower than the 30x to 50x coverage used for many contemporary whole-genome sequencing studies, Wang noted. But because the sperm cells are haploid, he explained, the same accuracy can be achieved at a lower depth of coverage.
"We don't need the high depth for the heterozygosity calls — at every single position there is only one allele," he said. "With 6x to 8x, the genotyping accuracy is actually better than it is for a 50x diploid whole-genome sequence."
The same lack of heterozygosity in the sperm cells can simplify things on the computational side too, he noted. For instance, the team tweaked the sequence alignment software somewhat, getting rid of sophisticated statistical models that most alignment tools use to deal with diploid data.
"If you use the alignment package as is, you would get a false result because 6x to 8x sequence depth for a diploid genome is regarded to be very shallow," Wang said.
On the genotyping side, he added, researchers wrote their own filtering algorithm to further narrow the results coming out of the Illumina Genome Studio software used to analyze array-based genotyping data.
Wang noted that it is also feasible to do recombination mapping using genotypes ascertained from whole-genome sequence data — something researchers did not do for the current study since deep sequence data was only available for a handful of individual cells.
In general, he explained, genotype information on at least 100 cells is needed to do recombination mapping with one centimorgan resolution, representing a 1 percent error rate, though having access to information on more cells is expected to further improve accuracy.
"With the development of single-cell amplification methods and the drop in sequencing and genotyping costs, that number will only go up," he said.
For the time being, genotyping, library preparation, and sequencing remain the largest expenses in this sort of single-cell study, Wang said, whereas the microfluidic step does not add significantly to the overall cost of such experiments.
The single-cell isolation and amplification strategy outlined in the Cell paper is expected to be compatible with any sequencing platform that is currently on the market, Wang said.
And going forward, the team hopes to apply this strategy not only to studies of other sperm cells, but also to analyses of diploid cell types, including circulating cancer cells and stem cells, according to Wang, who said the group is looking at the possibility of collaborating with other research groups within or outside of Stanford to do such research.