NEW YORK (GenomeWeb) – Building off research over the last several years that has sought to push the limits of analyzing the fetal genome by sequencing cell-free DNA in maternal plasma, researchers led by Dennis Lo at the Chinese University of Hong Kong have shown that by sequencing cell-free DNA to enough depth and incorporating bioinformatics filters, they can detect de novo fetal mutations.
In addition, the team's results, which were published this week in the Proceedings of the National Academy of the Sciences, shed some light on the biology of cell-free DNA, indicating that fragmentation of DNA in plasma is not random.
Jay Shendure, a professor of genome sciences at the University of Washington, who was not involved with the study, told GenomeWeb that the work was "an impressive demonstration of what might be possible" over the longer term. He said that while it would likely not have clinical implications in the near term, due to the cost and interpretation challenges, it represents what could be possible in decades.
He added that the research builds on studies that have been published over the years by Lo's group, Stephen Quake's Stanford University team, and his own lab at the University of Washington, on methods for sequencing the fetal genome from maternal plasma, getting haplotype information, and identifying inheritance patterns and de novo mutations.
Lo told GenomeWeb that although the current cost of such a test would be prohibitive, around $40,000, "with the continual reduction in sequencing costs, the cost of this technology will come down to a more realistic level."
One major advance in this study, Shendure said, is that while his team was able to show several years ago that it was possible to detect de novo fetal mutations, it did so with a very high false positive rate. In Lo's recent PNAS study, however, the researchers were able to dramatically reduce that false positive rate, demonstrating a sensitivity of 85 percent and a positive predictive value of 74 percent.
In the study, the researchers first developed their protocols on a plasma sample from a woman at 38 weeks of gestation. The sample was sequenced using Illumina's TruSeq PCR-free library prep to 270x coverage of a haploid human genome. In addition, the researchers sequenced DNA from maternal blood cells, paternal blood cells, and umbilical cord blood cells to between 40x and 50x coverage.
Next, they determined SNPs for which the fetus would be heterozygous, by looking at SNPs where both parents were homozygous but for different alleles. They identified more than 200,000 such SNPs and confirmed that they were able to detect the paternal allele in each case in the maternal plasma.
They then looked at the size differences between fetally derived DNA fragments and maternally derived DNA fragments, which they thought would be useful in determining the likelihood of a mutation being de novo in the fetal genome.
To look for de novo fetal mutations, the researchers first looked at SNVs present in the umbilical cord blood cells but absent in the parents, identifying 56 SNVs.
To see if they could then detect these 56 SNVs in the maternal plasma, they combined their deep sequencing with a bioinformatics protocol designed to filter out false positives. The protocol was able to identify high-quality base sequences and also took into account known biological characteristics of cell-free DNA. The sequencing alone was able to detect 54 of the 56 SNVs, but at a PPV of just .089 percent. Incorporating a filter based on realigning reads enabled 52 SNVs to be detected with a PPV of 35 percent. Two other filters that incorporated variant frequency and plasma DNA fragment size brought the detection rate to 48 out of 56 SNVs, or 85 percent, with a PPV of 74 percent.
Previous methods published by Quake and Shendure's groups to identify de novo fetal mutations relied on using haplotype information from the mother. The authors noted that haplotyping-based methods have a "limited resolution in which the maternal inheritance of the fetus can be determined," and so wanted to design a method that would not require haplotyping.
A second goal of the study was to look at patterns in DNA fragmentation, to determine whether certain genomic sites were preferentially located at the ends of the plasma DNA fragments. Using a PCR-free library step was critical for this analysis, since researchers did not have to worry about PCR duplicates.
By looking at both fetal and maternal DNA fragments that each contained a SNP—heterozygous in the fetal fragment and homozygous in the maternal fragment—the researchers used statistics to determine the probability of the molecules having been fragmented at the same spot relative to the SNP location. They then looked at the end positions of those fragments to see which were the same and which were different.
They found that 25 percent of the DNA fragments had at least one identical fragment sharing the same position at each end. If fragmentation had been random, only 1.45 percent of the molecules would share the same ends with one other molecule. In addition, in further analysis of plasma samples from 26 pregnant women, the researchers found that the plasma DNA fragments often had ends that seemed to be specific to the fetus or specific to the mother.
Lo said that one next step would be to continue studying this phenomenon and its implications. "I think there is a need to produce a catalog of such ending sites," he said.
The researchers also determined that there were size differences in maternal versus fetal DNA fragments, which could be used to filter de novo from inherited mutations. However, such filters would have to be applied on a case-by-case basis, since fragment size is also based on the fetal DNA concentration in the plasma.
Finally, the researchers tested their methods on a plasma sample from a woman in her second trimester whose fetus had been diagnosed with cardiofaciocutaneous syndrome. They sequenced the plasma DNA to 195x coverage and by using the series of filtering steps were able to identify the causative de novo BRAF mutation that had been previously identified on a microarray with a sensitivity of 81 percent and a PPV of 62 percent.
Lo said that the study has several important clinical implications, but also a number of challenges before such a technique could be used clinically. A number of monogenic disorders are caused by de novo mutations, so being able to detect those mutations prenatally could be important. In addition, there are known risk factors that predispose fetuses to developing de novo mutations, such as paternal age. Nevertheless, "gaps in our knowledge about the functions of many parts of the human genome do create challenges in our interpretation of the de novo mutations detected," he said. "I see this as a technology in progress."
Shendure agreed that cost and interpretation challenges would prevent such a technique from being used in the clinic in the near term, but said that eventually, it would be interpretable. Interpretation can be difficult even for alterations where penetrance is well understood and detection methods are highly accurate. And it is becoming even more difficult "as we're trying to expand the diagnostic range of NIPT to include smaller copy number events" with uncertain penetrance, he said.
Lo said that his lab plans to continue to work on the technology and would be particularly interested in studying the biology of the plasma DNA ends and their relevance in other fields aside from prenatal diagnostics. "It would be very interesting to look for plasma DNA ending sites that are associated with cancer," he said.