Building on a shotgun sequencing strategy employed in the detection of fetal trisomy from maternal plasma, Dennis Lo's team from the Chinese University of Hong Kong has shown that whole-genome sequencing of a cancer patient's blood sample can detect both copy number alterations and single nucleotide variants present in the patient's tumor.
In a proof-of-principle study published this month in Clinical Chemistry, the team demonstrated the method on four patients with hepatocellular carcinoma and one patient with both breast and ovarian tumors.
Other groups are developing methods to sequence circulating tumor DNA, such as a group from Cancer Research UK, which has developed a technique dubbed TAm-Seq (CSN 6/6/2012). However, said Lo, those other approaches have all been targeted, focusing on known cancer genes or hotspots, rather than sequencing the whole genome.
Lo told Clinical Sequencing News that his team is now working to validate the method in a larger cohort of patients of different cancer types. Eventually, the method could be used to "provide a genome-wide view of how a tumor evolves during the clinical course of a patient, [for example], how the genomic aberrations of tumors evolve in response to treatment or during relapse and progression."
Before it can be adopted clinically, though, Lo said that the diagnostic sensitivity and specificity will need to be validated on a large cohort and the cost will have to come down. Currently, he said, sequencing for one sample takes up four lanes of an Illumina HiSeq 2000, comparable in price to sequencing a whole tumor genome from a biopsy.
While the method is similar to work that Lo has published on sequencing fetal DNA from maternal plasma, it differs in that it measures both copy number changes and SNVs and the bioinformatics and data analysis is much more complex, said Lo.
"The spectrum of changes that one has to encompass for cancer detection by the bioinformatics algorithms is much larger than for the fetal work," he said. "The phenomenon of tumoral heterogeneity also creates a much greater degree of complexity for data analysis."
In the study, Lo's team collected blood samples from four hepatocellular carcinoma patients both before surgery to remove their tumors and one week after surgery, as well as DNA from each of the tumors. Samples from 16 healthy patients were also analyzed.
The samples were sequenced on the HiSeq 2000 and copy number alterations were also analyzed by Agilent and Affymetrix microarrays.
To assess copy number alterations, the team first divided the genome into 1-megabase sized bins, and mapped the sequence reads to the bins. The reads for each bin were counted up. Then the team developed a z-score statistic to determine if the plasma DNA in a specific bin increased or decreased when compared to the reference group of the 16 control individuals. Regions with z scores of less than three or greater than three were regarded as underrepresented or overrepresented, respectively.
Lo's team compared this approach with copy number analysis via microarray and found that the results were consistent. Next, they compared the approach with copy number analysis from the tumor tissue itself and found that the copy number alterations observed in the tumor tissue were consistent with the copy number alterations observed in the pre-surgery blood sample. Post-surgery, however, "such copy number aberrations disappeared almost completely" from the blood, the authors wrote.
The ability to detect copy number alterations from blood samples was dependent on a number of factors, including the amount of tumor DNA present in the blood as well as the type of alteration. For example, patient 1 had the largest tumor and over half of the DNA extracted from the blood sample was tumor DNA.
The other three patients had significantly less tumor DNA, comprising 5.6 percent, 4.3 percent, and 7.6 percent of the plasma, respectively.
As a result, nearly 100 percent of the copy number alterations could be detected from the blood of patient 1, while only between 10 percent and 70 percent of the copy number alterations could be detected from the other three patients.
The technique's sensitivity also varied depending on the type of alteration. Comparing 1-copy gains, 1-copy losses, and 2-copy gains, the technique could best identify 2-copy gains. For example, in patient 2, 70 percent of the 2-copy gain alterations were identified compared to only about 10 percent of the 1-copy gain alterations.
The authors wrote that the sensitivity could be improved by increased sequencing depth. Using computer simulations, they determined the required amount of sequencing to reach a detection rate of 95 percent for various tumor DNA concentrations. For a concentration of tumor DNA at 40 percent, to detect 1-copy gains and 2-copy gains, each 1-megabase bin would have to contain 180 and 800 molecules, respectively.
With a tumor DNA concentration of 10 percent, those numbers increase to 2,500 and 12,000 molecules per bin.
Lo added that further studies should also give a better idea of the average tumor DNA concentration in blood. From this study, the amount of tumor DNA present in the blood appears to correlate with tumor size, he said.
Additionally, said Lo, the GC bias of the Illumina sequencing platform reduces sensitivity. To call copy number variations, "you need to draw a threshold — three standard deviations from the normal. But if your baseline is varying, because of GC bias, for instance, you can't make calls as confidently," he said.
One way to correct for that would be to use single-molecule sequencing, he said, but currently the Pacific Biosciences platform does not provide enough throughput for this type of application.
To analyze the ability of sequencing to detect SNVs from plasma, the team sequenced the genome of each of the four patients' tumor genome from the biopsy, as well as a matched normal sample.
Then they compared the tumor SNVs with the SNVs detected from the plasma sample. The total number of tumor SNVs ranged from 1,334 to 3,170 in the four patient samples and the team was able to detect between 15 percent and 94 percent of them in the blood samples prior to surgery.
As with copy number variant detection, the SNV detection rate was correlated with the concentration of tumor DNA in the plasma sample. Aside from simply sequencing to greater depth, Lo said that improving the sequencing error rate itself would increase the assay's sensitivity because it would enable true variants to be distinguished from sequencing errors.
In order to assess SNVs from circulating tumor DNA, it is necessary to also have DNA from a tumor biopsy sample and matched normal sample, Lo said.
"Having the tumor sample beforehand would allow one to use plasma SNV detection by sequencing for monitoring the clinical progress of the patient," he said.
In cases where a tumor sample is not available, the copy number analysis portion can still be done by comparing copy number alterations from the plasma sample with a normal reference genome, he said.
Tumor Heterogeneity
Finally, the team tested the technique on a patient with both breast and ovarian tumors to see whether the plasma contained circulating tumor DNA from all tumors.
To do this, the researchers sequenced DNA from four regions in the two ovarian tumors, a sample from the breast tumor, and pre- and post-surgery plasma samples.
Looking at the different tumor samples, the team found that the breast tumor had a different pattern of copy number alterations from the ovarian tumors, while the four ovarian samples had very similar patterns.
Comparing the copy number alterations from the blood sample to those of the breast and ovarian tumor samples, the team determined that the breast cancer and ovarian cancers contributed 2.1 percent and 46 percent, respectively, to the plasma DNA.
Post-surgery, however, these numbers dropped to 1.3 percent and 0.66 percent, respectively.
Lo said that being able to assess tumor heterogeneity is a key advantage of the whole-genome approach over targeted approaches. Looking for just a handful of point mutations may not be representative of the patient's cancer, particularly if different clones proliferate or decline in response to treatment, he said.
Next Steps
Lo's team is now testing the protocol in larger numbers of clinical samples, including additional liver and breast and ovarian cancer samples, as well as cancer types that are particularly prevalent in Hong Kong, such as lung, nasal, and colorectal cancer.
The team will try to see whether the method is consistent across different tumor types, and is also working on strategies for reducing the cost and increasing the sensitivity.
He would also like to determine the importance of both the copy number portion and the SNV portion to determine whether they are "completely synergistic" or whether there "are conditions in which one is better than the other."
Ultimately, Lo said, the goal is to develop a clinical test that could be used to monitor patients' response to therapy, for instance. However, he stressed that a clinical test is still several years away.