NEW YORK (GenomeWeb) – A group of researchers from Stanford University has developed an error correction method for a circulating tumor DNA assay that enabled them to detect mutant alleles at a frequency of .004 percent.
The group described the method, called integrated digital error suppression (iDES), in a study published online today in Nature Biotechnology. The work builds on a ctDNA technique the Stanford team previously developed called CAPP-Seq, which Roche has since acquired.
Maxmilian Diehn, an author of the Nature Biotechnology study and an assistant professor of radiation oncology at Stanford, recently described CAPP-Seq and the basics of the error correction method at Cambridge Healthtech Institute's Molecular Medicine Tri-Conference in San Francisco, but provided more details on the strategy in this week's paper.
In the paper, the researchers wrote that the degree to which sensitivity of ctDNA assays can be improved is "ultimately limited by technical background."
The previous limit of detection for their CAPP-seq assay was .02 percent, but they said that at allele fractions below that, more than half of "sequenced genomic positions had artifacts."
To combat these errors, the team first designed a molecular barcoding strategy. Such schemes are used by many ctDNA assay developers to help reduce error rates by enabling sequenced calls to be traced back to their original DNA fragment. Both single-stranded and double-stranded DNA barcoding strategies exist, but each has their drawbacks, the authors wrote. Double-stranded DNA barcoding is better at reducing errors, but is less efficient than single-stranded DNA barcoding, so not as good for samples in which there is a limited quantity of ctDNA.
As such, they set out to design a hybrid strategy. First, they designed sequencing adapters that would enable both single- and double-stranded molecular barcoding. Each strand of a double-stranded molecule was first tagged with a four-base barcode, dubbed the index barcode. Next, they added two two-base barcodes adjacent to the ligating side of each adapter on both strands, called insert barcodes because they are sequenced as part of the main read of inserted DNA fragments. After sequencing, the complementary insert barcodes can be matched to reconstruct the original double-stranded DNA molecule.
For the second step, the Stanford team designed a computational tool to correct for systematic errors from sequencing or PCR. To do this, they first ran the CAPP-seq assay on samples from 12 healthy adults. The researchers reported that while there were recurrent background errors across all SNV classes, the most common errors were G-to-T transversions, with C-to-T and G-to-A errors also contributing.
They determined that the G-to-T errors occurred because of oxidative damage during the hybrid capture step. They designed a computational approach to suppress these errors, and when they combined this approach with the barcode strategy, they noticed a 15-fold reduction in error rates, which they confirmed across 30 healthy control samples and 142 samples from non-small cell lung cancer cell-free DNA samples.
The team also validated the assay on NSCLS samples. First, they evaluated its ability to detect hotspot EGFR mutations from a cohort of 41 patients with advanced NSCLC. The assay detected 142 EGFR variants in 88 plasma samples and all calls were verified in tumor biopsy samples. Moreover, the assay never had any false positive EGFR calls in samples known to be wild type.
Next, they assessed the technical limitations of the CAPP-seq assay with iDES on reference cell lines with known variants spanning a range of frequencies. They created a blend of the reference cell lines that contained variants at allele frequencies of .05 percent to 1.6 percent. They examined both the barcoding strategy and computational error correction steps separately, as well as together, determining that the two methods were complementary and produced better results than each method separately. They also determined that the theoretical limit of detection for the iDES CAPP-Seq assay could be 2.5 molecules in 1 million, or .00025 percent.
Finally, to test the actual limit of detection for monitoring disease in patients, they used the assay to monitor mutations in plasma samples from 30 NSCLC patients whose tumors had previously been genotyped.
In pre-treatment samples, the assay could detect ctDNA in 93 percent of the patients, including all three stage I tumors. In addition, the researchers found that they were able to detect a mutation at a frequency of .004 percent in a patient prior to clinical relapse. "To our knowledge, this is the lowest amount of ctDNA detected by deep sequencing in any NSCLC patient to date," the authors wrote.