SAN FRANCISCO (GenomeWeb) – Circulating tumor DNA holds great promise as a way to molecularly profile cancer patients' tumors, to monitor their response to therapy, and to identify markers of residual or recurrent disease as well as drug resistance. One difficulty with ctDNA, however, is that any given somatic mutation is present at a very low frequency — often making true mutations difficult to distinguish from errors.
At a Cambridge Healthtech Institute symposium following the Molecular Medicine Tri-Conference here this week, a number of researchers described strategies they have developed to be able to distinguish lower-frequency mutations in ctDNA from sequencing and PCR errors.
The problem of trying to detect very low-frequency variants in background that is mostly wild-type is not new in the liquid biopsy world. As such, a number of companies developing such assays have implemented their own proprietary methods to enable them to distinguish variants from errors.
Trovagene, for instance, uses a proprietary method that enriches specifically for mutant alleles, as well as an extraction method that enables detection of ctDNA from urine. Guardant Health uses a barcoding strategy in its sample prep method, adding oligonoucleotide barcodes to both strands of each individual DNA fragment.
At the Tri-Con symposium, Maximilian Diehn, an assistant professor of radiation oncology at Stanford University, described a method he developed with Ash Alizadeh, called CAPP-seq. The team originally published the method in Nature Medicine, and spun out a company called CAPP Medical that has since been acquired by Roche.
In the 2014 Nature Medicine study, the team assessed the CAPP-seq assay on 40 samples from 13 patients with non-small cell lung cancer, demonstrating that they could detect mutations down to a 0.02 percent frequency.
At this week's conference, Diehn described an improved version of CAPP-seq that incorporates error correction to be able to detect mutations present at frequencies of 0.001 percent.
In order to lower the detection limit, Diehn said the researchers implemented a number of improvements.
The main limitations in sensitivity are errors from PCR in the sample prep steps and errors in the sequencing itself. To try and fix this problem, the team first wanted to see whether there were systematic errors. So they ran the panel on plasma from healthy donors and identified and plotted the errors. They noticed that the errors were "very biased," Diehn said.
To correct for these errors, they first added a molecular barcoding step, tagging each cell-free DNA fragment with a unique identifier. That way, when mutations are called they can be tracked back to the original DNA fragment. "That helps suppress some errors," Diehn said, but did not get rid of all of them. So they next designed a statistical method called integrated digital error suppression (iDES), details of which he said would be published soon in a journal.
They then tested the assay cell lines, spiking in mutations at various frequencies. The two steps synergistically reduced errors and enabled the detection of mutations down to 0.001 percent frequency.
Diehn said that the team is also validating the method on samples from patients for whom they do not have tumor biopsy. In the earlier study, the researchers would first run the assay on tumor biopsy samples to identify the set of mutations they wanted to track in subsequent blood draws. But, recognizing that not all patients have available tumor biopsies, one goal has been to demonstrate that the test would work from a treatment-naïve blood sample.
By comparing samples taken from the blood to tumors of the same patient prior to starting therapy, they were able to show that the assay was "highly concordant," Diehn said. In addition, he noted, the ctDNA assay provided a good assessment of tumor burden.
The group has also validated the assay on other tumor types aside from non-small cell lung cancer, including breast, ovarian, diffuse B-cell lymphoma, pancreatic cancer, and others, Diehn said.
Meantime, a group at Yale University is also using molecular barcoding as part of a strategy to identify true variants from errors when looking for low-frequency mutations in ctDNA.
Abhijit Patel, an assistant professor at Yale, said his group has designed an NGS-based assay that consists of 50 amplicons. The team uses two methods of error correction. First, he said, they take advantage of paired-end sequencing and the fact that ctDNA fragments tend to be relatively short, averaging 175 bases. The amplicons are designed such that when they are sequenced, the hotspot region is sequenced in both the forward and the reverse direction. "The chances of having the same error on both is very low," Patel said.
In addition, he said the group also uses molecular barcoding to enable them to distinguish errors caused by PCR from true variants. "If you have a true mutation in a template molecule, you'd expect to see multiple copies of that mutation associated with that molecular barcode," Patel said. "But if there is an error that arises during PCR, you'd expect to see several different barcodes associated with that error."
The assay has an error rate of less than one in 100,000 bases. Per milliliter of input plasma, the assay has a sequencing cost of $20, Patel said, and they are able to multiplex 100 samples on one lane of the Illumina HiSeq instrument.
Now, he said, the group is working on method that is able to use just one tube and perform PCR and barcoding all in the single tube. To do this, he said they are working with David Zhang and his lab at Rice University, who has designed a technology called a toehold probe. Zhang described the technology, which improves single-base discrimination of hybridization reactions, in a 2012 Nature Chemistry study.
Patel said the group is now working to validate the technology on clinical samples, and has so far tested more than 1,500 samples. He described one case of a 72-year old man with metastatic lung adenocarcinoma with an EGFR mutation. Ultimately the patient continued to progress and developed metastasis on the liver. Signs of the metastasis were seen in the ctDNA months before a CT scan could detect the tumor.