Genentech researchers have developed a targeted sequencing panel that they are using to analyze archived clinical samples and clinical trials in order to identify biomarkers associated with disease progression, drug response, and resistance.
The method combines microfluidic multiplex PCR-based target enrichment of 88 cancer-related genes with next-generation sequencing on Illumina's platform and was detailed in a recent publication in Clinical Cancer Research.
The researchers are now using the technique as a "biomarker discovery tool," Yulei Wang, a scientist in Genentech's oncology biomarker development division told Clinical Sequencing News. "We have implemented the method to do translational studies from tissues from clinical trials … to maximize what we learn from these trials," she said. "It's a tool to look for clues — patterns of disease progression, resistance patterns, drug response."
Wang said that the team chose to develop its own assay as opposed to using a commercially developed target enrichment technology because they wanted to have a better understanding of the ability to analyze poor-quality clinical samples from formalin-fixed paraffin-embedded tissue.
Aside from the panel itself, the team also developed a "ruler" assay to predict whether a given FFPE sample would yield a reliable result.
Wang said her team works on oncology biomarker development, where "the unique challenge is the sample. We're working directly on patient tissue samples and 90 percent are FFPE samples. There is a lot of DNA degradation and modification, and so, we were thinking about what is the best approach to enable a reliable genetic mutational profile for these challenging tissues."
To design the panel, the team selected 88 genes, including ones with known clinical actionability (defined as those that encode drug targets or have diagnostic or prognostic utility), other genes that are frequently mutated in cancers, and genes that the Genentech team is interested in studying further as potential biomarkers, Wang said.
She added that they went with a PCR-based target enrichment strategy as opposed to hybridization capture because PCR is typically is more amenable to working with limited sample.
For 15 tumor suppressor genes, the researchers designed amplicons to target all exons and splice junctions. For the other 72 genes, the team designed amplicons to target only hotspot mutations. They designed a total of 963 amplicons that covered around 150kb of sequence, including 268 exons, 340 mutation hotspots, and more than 10,000 COSMIC-annotated mutations.
Because FFPE samples are highly degraded, the team created short amplicons — 89 percent were shorter than 150 bp and all were under 200 bp.
Next, they used Fluidigm's microfluidic Access Array System to apply 48 PCR reactions to 48 samples. Wang said the team chose to use a microfluidics approach because of the ability to enable automation and high throughput. They divided the amplicons into two panels with 10 to 12 PCR primer pairs per reaction in order to simultaneously amplify 480 or 483 distinct regions on a single Access Array.
The researchers pooled the PCR products from each sample and barcoded them for sequencing.
To gauge the panel's accuracy, the Genentech team tested it on 66 cancer cell lines with known mutations.
Using 50 nanograms of starting DNA, each sample generated an average of 1.4 million reads, with 98 percent mapping to the region of interest.
Next, they wanted to test the panel's ability to detect low-frequency mutations. The team created dilutions of seven cancer cell lines with mutually exclusive somatic mutations with variants ranging from 40 percent down to 0.2 percent frequency and also compared the assay to Sequenom's mass spectrometry-based MassArray OncoCarta Panel and a multiplexed PCR technology. The average detection limit for Genentech's MMP-seq assay was 1.8 percent, compared to 6.6 percent for the MassArray panel and 3.2 percent for the PCR-based assay.
Further characterization of the MMP-seq assay in 47 cancer cell lines also represented in COSMIC found that MMP-seq detected all 97 previously reported SNVs and 22 of 24 previously reported small indels. In addition, the team validated 25 of 27 novel and potentially deleterious SNVs not annotated in COSMIC. "Several of these may potentially impact pathway activation and drug sensitivity," the authors wrote.
Once the researchers established analytic performance of the MMP-seq assay, they wanted to test its ability to analyze clinical samples from FFPE tissue, so they ran the panel on 17 clinical samples for which there was both FFPE and fresh frozen tissue available. Additionally, the team developed a so-called ruler assay to help judge the quality of a given FFPE sample.
"The ruler assay uses a tiny amount of material and is able to suggest whether you're going to have success with any given sample," Richard Bourgon, a bioinformatician and computational biologist at Genentech, told CSN. "It gives you a nice way to use little of the precious sample to get a sense of whether it's worth pursuing, and if so, [whether you can] treat it as a fresh sample or … approach with caution."
The ruler assay estimates the number of functional DNA copies available for target enrichment. It is based on qPCR of the TRAK2 locus, which the team chose because its amplicon length of 149 bp is consistent with the MMP-seq library, and also because copy number variation of TRAK2 is rare.
The ruler assay found a wide range of estimated functional DNA copies — from 200 to 5,800 copies in each 150-ng sample. It was also highly predictive of concordance between the matched fresh frozen and FFPE samples. For instance, in samples with low estimates of functional DNA copies, a large number of false positives were found in the FFPE samples, including both low frequency and high frequency calls.
Nevertheless, even in samples deemed low quality by the ruler test, the MMP-seq assay could still detect well-characterized hotspot mutations. In 116 hotspots across 11 genes, there was "near-perfect concordance" between the fresh frozen and FFPE samples across all 17 patients, the authors wrote.
Examining the source of false positives in the low-quality FFPE samples, the researchers discovered that the main reason was due to deamination, a known problem in FFPE samples. Strategies exist for repairing deaminated DNA, such as pretreatment with uracil-DNA glycosylase, which removes uracil-containing deaminated molecules. The researchers found that pretreatment reduced false positives by 77 percent for C to T changes and by 94 percent for G to A changes.
Finally, the researchers tested the assay on endometrial cancer FFPE samples. Currently, there are drugs in early clinical development that target genes within the PI3K/AKT/mTOR pathway, according to the researchers, and they wanted to see whether the MMP-seq panel could detect mutations in those genes, which included PTEN, PIK3CA, AKT1, KRAS, PIK3R1, and PIK3R2. From 73 samples that were of sufficient quality, they generated on average 1.2 million reads per sample with 98 percent of reads on target and obtained a coverage uniformity of 89 percent. Additionally, they compared their approach with multiplex PCR genotyping at 81 SNVs.
The next-gen sequencing panel identified 54 out of 55 variants identified by PCR. Sequencing also identified an additional 27 variants that were not detected by PCR. Most of those variants were present at a low frequency — between 2 percent and 5 percent.
Moving forward, Wang said that the team plans to use the assay for biomarker discovery. Genentech does next-gen sequencing studies both in-house and also outsources. The company has sequencing systems from all the major vendors, Bourgon said, including Illumina MiSeqs and HiSeqs, Life Technologies' Ion Torrent, and Roche's 454.
"Our sequencing core is applying a hybrid model" of both in-house and outsourced sequencing, Bourgon said. "The model our sequencing core has found to be effective is to start ourselves with the new technology so we understand where the quality points for output [are]." Understanding the various systems and their performance characteristics makes it easier to both negotiate with vendors on price and to assess the data when it is returned. However, he added, no matter where the sequencing is done, bioinformatics is always done in house.
The bioinformatics pipeline is "tailored and optimized for clinical sequencing, and we know all the parameters inside and out and can constantly be improving them," Wang said.