This story has been updated to clarify that samples sent for sequencing by Ambry Genetics in the Guardant study were sequenced using non-clinical, research-grade NGS.
NEW YORK (GenomeWeb) – Close to a year and a half into commercially offering its targeted sequencing-based liquid biopsy test, Guardant Health has published the first comprehensive report on its technology's analytical and clinical validity.
The study, which appeared online last Friday in PLoS One, demonstrated near perfect analytic specificity of 99.9999 percent and sensitivity of 100 percent in experiments comparing the initial 54-gene version of its Guardant360 test with tissue sequencing results.
In addition to analytic evaluations, the company also reported that its test showed a clinical sensitivity of 85 percent based on a comparison of 165 matched tumor and blood samples from a small multicenter study.
Finally, researchers shared additional data on the first 1,000 patients Guardant Health sequenced commercially in its first five or so months of operations, showing that the majority were found to have an actionable variant of some kind, either linked to an approved drug, or making the patient in question eligible for a clinical trial.
Helmy Eltoukhy, Guardant Health's CEO, told GenomeWeb in an email that this study, along with several other recent publications, provides crucial evidence of Guardant360's validity and utility for potential users.
"In addition," he said, "these studies help payers understand that our blood biopsy brings patients to the highest standard of care, without the costs and complications of a tissue biopsy. We think the health economic arguments of non-invasive testing are self-evident and strong."
Guardant had not made public much detail about its approach to liquid biopsy sequencing, which it claims creates a much higher resistance to the errors and bias that sequencing introduces than other technologies. But in the company's recent study, its researchers provided a look at the method the company calls "digital sequencing," which involves library preparation using oligonucleotide barcoding of each strand in every extracted DNA fragment.
More specifically, the study authors wrote, Guardant's digital sample preparation uses non-unique heptamers ligated to each half of individual double-stranded cell-free DNA, generating a duplex library. Following this preparation, Guardant then sequences a panel of genes — 54 in the first panel used in the validation study, and now up to 70 genes in the panel the company is currently offering commercially — using the Illumina HiSeq 2500.
Post-sequencing, the company's analysis methods consider both strands of each sequenced cfDNA molecule in a process authors described as being akin to building a separate noise model for each of the 78,000 bases covered by the initial 54-gene panel. Software compares the two complementary strands to ascertain whether either contains errors from sequencing, library prep, or DNA damage during sample processing.
In Guardant's new study, researchers from the company demonstrated — via sample spiking experiments and comparison to matched tissue exome sequencing data — that Guardant360 was sensitive down to 0.01 percent mutant allele fraction.
In liquid biopsy, the ability to detect mutations at extremely low frequency amidst the vast background of normal DNA that circulates in the blood is an absolute necessity. High specificity is also important for successful analysis of circulating tumor DNA, especially as companies like Guardant and others work towards adapting their technology for not just testing in known cancer patients, but also early detection of cancer in asymptomatic or otherwise healthy people.
In its analytic validation, the company sent DNA from 20 healthy donors to an outside reference lab (Ambry Genetics) for research-grade whole-exome sequencing, and then spiked 10 nanograms of matched cfDNA from each of those same subjects at a concentration of five percent into another unique cfDNA sample for processing using the Guardant sequencing method.
Compared to the WES results across the 54 genes that Guardant's test covered at the time, Guardant360 correctly identified all of 365 WES variants with one additional false positive relative to the WES results. According to the study authors, with a cumulative targeted region of 1.56 million base pairs across the 20 tested samples, this corresponds to an analytical specificity of 99.9999 percent and sensitivity of 100 percent.
For context, the study authors cited published specificities for other liquid biopsy sequencing methods ranging from 96 percent to 99.2 percent specificity.
Authors of Guardant's validation study stressed that although the company calculated a less than perfect specificity, the actual validity of that measurement depends on the accuracy of the reference standard used, in this case whole-exome sequencing.
In other words, it's possible that Guardant's one false positive could have actually represented an NGS false-negative. To address this question, the researchers sent out a sample for Sanger sequencing, which appeared to confirm the Guardant test results — a SNP in the HRAS gene — over the tissue NGS results. Thus, the researchers amended their judgement of the assay's analytical specificity to "greater than 99.9999 percent."
In addition to speaking to its tests' analytical power and reporting on assay repeatability and robustness, the Guardant study also included data from a small multicenter clinical validation, in which the company analyzed 510 plasma samples from stage III and IV non-hematologic cancer patients from five different centers.
Of 510 total samples, close to 99 percent were sequenced successfully. Of these, 86 percent were positive for a genomic alteration of some kind.
A subset of 165 patients in the 510-subject validation also had matched tissue samples, and the researchers used these to compare the performance of Guardant's test with tissue sequencing. Compared to tissue sequencing results, the authors reported, the clinical sensitivity of Guardant360 was 85 percent, and its specificity was more than 99 percent.
"The sensitivity was 85 percent when you look at tumor as gold standard," AmirAli Talasaz, Guardant's president and COO, told GenomeWeb last week. "But, in our analysis we also flipped it around so if you look at cell-free DNA as the standard, the accuracy of tissue sequencing is 80 percent."
Though more data will be needed to establish definitively what the standard should be in comparing tumor tissue and circulating DNA results, Talasaz said that Guardant's perspective is that given the analytic power it has demonstrated, it can view its results with high confidence.
Really, the gold standard is the clinical response of the patient
"Really, the gold standard is the clinical response of the patient," he added.
For example, he said, a patient with gastric cancer who was negative for HER2 amplification based on two separate commercial tissue sequencing assays but positive using Guardant360 was later found to have HER2 overexperssion based on IHC and responded well to HER2-targeted therapy.
In lung cancer, the company has also seen several cases where patients with tissue sequencing showing an activating EGFR mutation have become resistant to EGFR-targeted therapy, and Guardant's ctDNA analysis has identified new resistance markers like T790M, which were not present in a patients' initial tumor sample.
"We've then seen patients matched to third-generation EGFR TKIs who are showing good response," Talasaz said.
Guardant investigators also reported in their study last week on the first 1,000 consecutive samples that the company processed commercially in its CLIA lab. Among 1,000 submitted samples, the study authors wrote that Guardant successfully sequenced 998, representing a 99.8 percent assay success rate. All but four of these cases had results that could be reported back to patients. The four that could not were due to confounding of the analysis by the presence of fetal DNA or transplant donor DNA.
The team reported that 78 percent of the sequenced samples tested positive for at least one genomic alteration, and of these, 76 percent were deemed "actionable." About 12 percent of these had an FDA-approved therapy for the specific mutation and cancer type in question. Another 52 percent were linked to a matched therapy but in a different cancer type. The rest had a genomic alteration that made a patient potentially eligible for a clinical trial.
Interestingly, along with validation results, the study also included some data on healthy subjects sequenced using the company's panel as part of its technology development and then its clinical testing operation.
In 143 healthy controls sequenced during commercial patient testing, none showed any somatic mutations in the company's 54-gene panel, the study authors reported. But among another 79 healthy control samples tested during the Guardant360 tech development process, the company did find a single TP53 mutation in a heavy smoker with no history of cancer.
This result speaks obliquely to the potential of Guardant's digital sequencing approach to achieve a high enough sensitivity to detect cancer-associated mutations in people without symptoms of a malignancy.
Like many other companies in the liquid biopsy space, Guardant has said it is looking toward adapting its technology for testing not only patients with known cancer, but also for screening or early cancer detection in at-risk but otherwise healthy individuals.
Moving forward, Talasaz said Guardant is continuing to track patient outcomes as its cohort of commercially tested cases grows.
Recently, the company shared additional data on 6,500 commercially tested samples, he said, and at this point, Guardant's run rate has risen north of 20,000 samples per year.
The company is also continuing to build collaborations with outside academic groups. One, from the Samsung Medical Center in Korea, recently published its own external study on the correlation of Guardant360 results with tumor sequencing in the journal Oncotarget.
"We reported Guardant360 results to them but we were blind to the tumor status," Talasaz said.
The Samsung team then compared ctDNA and matched tumor samples for 61 metastatic cancer patients and another 14 stage II colorectal cancer patients. Overall, the concordance rate of ctDNA results with tumor results was 85.9 percent, the authors wrote — very close to the results in Guardant's own validation study.
Talasaz also highlighted ongoing work with MD Anderson, as well as Guardant's participation in the National Cancer Institute-funded SWOG study as sources for more data in the future on the impact of Guardant360 on patient outcomes.
"That to us is going to be really key to making an impact in cancer care, showing that ctDNA assays can make a benefit for patients," he said.