NEW YORK (GenomeWeb) – Researchers from the Dana-Farber Cancer Institute, Broad Institute, and other collaborators have designed a next-generation sequencing workflow especially suited for analyzing tumor genomes from formalin-fixed paraffin-embedded tissue.
In a study published this week in Nature Medicine, the team demonstrated that its exome sequencing protocol yields equivalent results when applied to DNA extracted from FFPE tissue or fresh frozen samples. Additionally, the researchers developed an algorithm specifically suited to ranking cancer mutations by clinical relevance. When applied prospectively, the protocol identified clinically relevant alterations in 15 out of 16 patients.
The goal was to "come up with a soup-to-nuts way of doing exome sequencing to inform clinical care in oncology," Eliezer Van Allen, co-first author of the paper and a post-doctoral research fellow at the Broad Institute, told In Sequence.
The first question that had to be tackled was whether the DNA extraction and subsequent sample prep steps would perform as robustly on FFPE tissue as on fresh frozen samples, Van Allen said. "The gold standard for genomics in oncology is frozen tumor samples," he said. "But that's not the way most clinical samples are stored."
DNA from FFPE samples is often more degraded than DNA from fresh frozen samples due to the fixing process. In addition, there is much less DNA, so nucleic acid extraction methods and library preparation protocols must be as efficient as possible.
The researchers used a slightly modified protocol from one previously developed at the Broad Institute to automate library construction for exome sequencing. For library construction, they reduced initial genomic DNA input from three micrograms to 10 to 100 nanograms. Additionally, they replaced Illumina paired-end adapters with Integrated DNA Technologies' palindromic forked adapaters with 8-base molecular barcodes.
They used Agilent's SureSelect in-solution exome capture kit with a few modifications to the hybridization reaction to normalize the concentrations of the libraries.
Next, they sequenced the libraries on the Illumina HiSeq using 2x76 bp runs.
According to Van Allen, while there were a "few technical details in extracting the DNA" that had to be incorporated to better handle FFPE samples, "practically speaking, the protocols [for FFPE and fresh frozen tissue] are not dramatically different, and that was a helpful thing we learned."
Van Allen acknowledged that while the fixing process does introduce some artifacts into the DNA, in the majority of cases in which researchers have had serious problems with FFPE samples, the samples have been very old. But, in the current study, the samples were no more than seven years old, he said.
In a comparison of 99 FFPE samples to 768 non-FFPE samples, the researchers found there were no significant differences in standard whole exome sequencing metrics, which they defined as achieving greater than 80 percent of the target with at least 20x coverage, and over 100x average coverage across the entire exome. Additionally, they were able to achieve a turnaround time of 20 days for samples received as FFPE tissue blocks.
"Even when we used only 1 ng of input DNA we saw equivalent results with DNA derived from FFPE and non-FFPE tissue," the authors wrote. However, for such low inputs, "a disproportionate amount of additional sequencing was required."
Van Allen said the study was "reassuring" that "a reasonable FFPE sample from an average clinical case approximates the results from a frozen sample."
To test whether FFPE samples yield the same mutation calls as fresh frozen tissue, the researchers examined exome data from 11 lung adenocarcinomas for which tumor and normal tissue were available from matched FFPE and frozen tissue and found that the data was more than 90 percent concordant. They attributed discordant results primarily to regions that were poorly covered.
Finally, the team applied an algorithm, dubbed PHIAL, especially designed for interpreting results from tumor sequencing.
PHIAL answers a "different question than we're used to asking in the research world," Van Allen said. It's "oriented toward one patient in time."
In developing PHIAL, the team first curated the literature and existing databases to develop TARGET, a database consisting of 121 genes with prognostic, diagnostic, or therapeutic implications. Those were then integrated with existing open-source resources to create rules that sorted each variant by clinical and biological relevance, linked genes with additional biologically significant pathways and gene sets, and denoted variants of uncertain significance. The algorithm also takes into account specific situations, Van Allen said, for instance two hits in the same pathway elevates both of those mutations. "That particular node would be ranked over others," he said.
To test the entire protocol prospectively, the researchers performed exome sequencing and applied the PHIAL algorithm to 16 patients with advanced cancers and identified 41 clinically relevant variants in 15 patients. Around 46 percent of those were based on preclinical evidence for the association with response or resistance to an approved therapy. Additionally, the team identified multiple unexpected findings. For example, they found a CRKL amplification in a patient with metastatic urothelial carcinoma, which has been predicted to confer resistance to EGFR inhibitors and sensitivity to Src inhibitors, but had never been seen in urothelial cancers.
For one patient, the results led to a change in treatment. A KRAS mutation, which had not been found with standard clinical testing, was found in a patient with metastatic non-small cell lung cancer. The patient had been progressing on chemotherapy, so the oncologist decided to enroll the patient in a phase I clinical trial of a CDK4 inhibitor on the basis of preclinical data pointing to a relationship between activated KRAS and CDK4. The patient stabilized. "Of note, this was the patient's best and only clinical response to any cancer-directed therapy," the authors wrote.
Van Allen noted that one important goal was to start developing a scheme to standardize the classification of cancer-related somatic variants. The American College of Medical Genetics issued guidelines recently for classifying germline variants as pathogenic or benign, but "for somatic mutations, there is no common language and no universally agreed upon way of classifying variants," Van Allen said. "That lack of a common language is a challenge we're trying to overcome."
The PHIAL scheme takes into consideration whether a given variant is associated with an approved drug or investigational compound, whether there is preclinical work supporting it as a therapeutic target, or whether the evidence is simply from computational models.
Currently, said Van Allen, the term actionable can have different meanings to different people. "There's a lot of nuance," he said. "Giving more detail about what that actually means is important for rolling [exome sequencing] out in the clinic."
Van Allen added that the paper attempted to present its comprehensive version of clinical cancer exome sequencing in order to start a discussion in the community about developing a common language and standardizing the process. Now, he said, the researchers are applying it to larger patient cohorts. For instance, it is the protocol being used in the CanSeq study that is evaluating how oncologists use exome data. Levi Garraway, senior author of the Nature Medicine study, is heading CanSeq, which is also one of the flagship projects of the newly created Joint Center for Cancer Precision Medicine that leverages resources from the Broad Institute, Dana-Farber, Brigham and Women's Hospital, and the Boston Children's Hospital.