SAN FRANCISCO (GenomeWeb) – A team from the Institute of Cancer Research, London has developed a cell-free DNA assay to detect low-frequency mutations in colorectal cancer. It plans to use the assay initially for translational studies but also aims to develop a clinical version of the test that might also be applicable to lung and and other cancers.
The assay makes use of hybrid-capture reagents, molecular barcodes, and an error correction bioinformatics scheme. In a study published in Clinical Chemistry last month, the researchers reported that it detected 100 percent of known variants down to a variant allele frequency of 0.15 percent.
Marco Gerlinger, who heads the translational oncogenomics team at ICR and is a senior author of the study, said that although there are a number of commercially available cfDNA assays, the group decided to develop its own because existing assays are limited in terms of target genes and customizability.
"We're interested in discovering resistance mechanisms and new drivers of metastatic disease, so we needed the option of changing around what we want to include in our sequencing assay," he said. While commercially available assays that make use of amplicon technology are customizable, he said, the team also wanted the ability to look at genome-wide copy number alterations, which can be more challenging with amplicon-based assays than with hybrid capture-based ones.
While amplicons are great for "very focal analysis," he said, "it can be difficult to target longer exons," because primers have to be tiled across overlapping regions in a way that they don't interfere with each other. With hybrid capture, however, it is possible to take advantage of off target reads, which in many situations are " a nuisance," and use them to "generate a copy number profile of the entire genome," he said.
Gerlinger's team primarily focuses on gastrointestinal cancers, which are highly aberrant at the copy number level, so it is important to be able to see those changes, he said.
For the study, the researchers designed a 32-gene assay, 163.3-kilobase pairs in size, using Agilent SureSelectXT HS technology, to which they had early access. The reagent kit incorporates 10-base molecular barcodes.
To calculate the sensitivity and specificity of the assay, the team created a mixture of cfDNA with 16 known SNPs within the targeted region at various concentrations. At 0.15 percent variant allele frequency, all 16 SNPs were detected. At frequencies of .075 percent and .0375 percent, 14 and 11 of the 16 SNPs were detected, respectively.
Next, in order to improve their analysis, the researchers developed a bioinformatics tool, called DuplexCaller, which uses the fact that a mutation should be seen in both the forward and reverse strand of a paired-end read to improve the accuracy of variant calling.
"The molecular barcodes were used as the primary method to remove sequencing and PCR errors," Gerlinger said. The DuplexCaller, "trying to identify the two DNA strands that were initially paired together, gives us an additional level of confidence and lets us remove the remaining false positives."
The group then tested the assay on cfDNA from 28 patients with metastatic colorectal cancer who also had a tumor biopsy sequenced. The cfDNA assay detected 80 of 91, or 88 percent, of the mutations that had been found by tumor sequencing. The 11 mutations that were not called by cfDNA were from three patients. Five of those mutations were found with manual inspection of the data, but at allele frequencies too low to pass the bioinformatics filter. For one case, there was enough remaining cfDNA to see whether one of the missed mutations, a BRAF mutation, could be found using droplet digital PCR. The ddPCR assay also did not detect the BRAF mutation, suggesting that it was not missed because the sequencing assay had failed.
The cfDNA sequencing assay also detected mutations in genes that had not previously been analyzed by tumor sequencing. For instance, in four cases that had previously been analyzed via a 5-gene amplicon-based assay, the cfDNA assay detected mutations in the APC gene, a well-known tumor suppressor gene. In addition, the cfDNA assay detected mutations in the FBXW7, CTNNB1, TCF7L2, ATM, and SMAD4 genes. Eleven out of 13 mutations detected resulted in protein changes and have been reported in the COSMIC database.
Finally, the group evaluated the instances where cfDNA detected a mutation in a gene that was analyzed with the tissue sequencing assay but not detected. There were 22 such mutations, found in TP53, ATM, PIK3CA, SMAD4, KRAS, FBXW7, and TCF7L2. The authors estimated that 12 of those mutations were cancer drivers, while two of the TP53 mutations were clonal hematopoiesis mutations.
In one patient, an activating KRAS mutation was found in cfDNA but not the tissue biopsy. The mutation was confirmed in the cfDNA via ddPCR. That patient had received anti-EGFR therapy before blood collection and the authors hypothesized that the KRAS mutation could have developed as a result of acquired resistance. For the discordant PIK3CA mutations, the authors suggested that the discordance could be due to tumor heterogeneity. In two such cases, the individuals went on to undergo "parallel evolution events," according to the authors, with additional PIKCA mutations in both the tumor tissue and the cfDNA. The discordant mutations found in the tumor suppressor ATM could not be interpreted and were thought to be potentially false positives. Those mutations were identified in cfDNA at a very low frequency, an average of 0.17 percent, and only two of the mutations had been previously cataloged.
Aside from detecting point mutations, Gerlinger said, one goal of the assay was to use the off-target reads to get a copy number profile. The team was able to generate copy number profiles for 20 out of the 28 samples and found one targetable amplification in the ERRB2 gene that was also detected in the matched tumor.
Gerlinger said that going forward, the lab plans to use the cfDNA assay as a "key research tool in early-stage translational trials for discovery." The team focused on designing an assay that maximized the number of genes that would be relevant for such trials, while also being limited enough to keep costs low.
He said two main goals are to use the assay to determine which mutations are driving metastasis and drug resistance. "We want to use this to detect the relative size of the drug resistant subclones. To do that, we need to combine copy number and variant frequency analysis to infer what percentage of cancer cells at resistance harbor resistance-conferring mutations."
In addition, he said, other researchers in the clinical sequencing lab are evaluating the assay to see whether it can be used for routine clinical testing. For clinical testing, a cfDNA assay would likely include not just genes related to colorectal cancer but also genes relevant for other cancers where liquid biopsy assays have shown to be effective and for which biomarkers exist, such as a lung cancer.
"It would be a major advantage to run such an assay locally," he said, as opposed to ordering testing through a commercial lab.
On the technology development side, Gerlinger said that his ICR team is going in the opposite direction — seeing how broad cfDNA testing can be. "We want to see how far we can push the technology," he said. The group recently completed a pilot study looking at low-coverage exome sequencing of cfDNA. "We're analyzing those results now and looking at where we need to improve the technology," he said.