NEW YORK – As companies and investigators push closer to validating liquid biopsy tests for cancer screening and early detection, approaches relying on epigenetic and DNA fragment patterns are leading the way, according to both commercial and academic players.
At the American Association for Cancer Research Advances in Liquid Biopsy meeting last month, for example, presenters included commercial firm Grail, which highlighted some of its most recent data on a targeted methylation-based pan-cancer detection assay.
Other recent scientific meetings have also featured new data from firms like Freenome and the Laboratory for Advanced Medicine, which are also harnessing epigenetic signals but have taken a more targeted tack in focusing on single cancer types.
Techniques have broadened to include not only methylation-specific signatures like Grail's or LAM's, but also fragmentomic methods that differentiate cancers from controls via patterns in how DNA appears to have been broken apart. Still others combine one or both with more traditional genomic targets like SNPs and copy number changes.
At the AACR meeting in January, researchers from various academic institutions also discussed advances they have been making in driving up the sensitivity of epigenetic cancer detection algorithms that use both DNA methylation and fragmentation signals.
Kicking off the meeting, Dennis Lo shared data on how the addition of DNA fragment size analysis and methylation signals allowed him to significantly improve the sensitivity of his Epstein-Barr virus nasopharyngeal cancer detection test.
Johns Hopkins oncology professor Viktor Velculescu also presented on work his lab initially published last year to develop a method for cancer detection using low-depth genomic sequencing to identify differences in DNA fragment length, fragment-length variability, and nucleosomal positioning that are specific to cancers versus controls.
In that earlier study, the group analyzed the fragmentation profiles of 236 patients with breast, colorectal, lung, ovarian, pancreatic, gastric, or bile duct cancer and 245 healthy individuals. When combined with mutation detection, they could detect 91 percent of the cancer cases. They also found that DNA fragmentation profiles could be used to narrow down the tissue of origin of the cancers in about 75 percent of cases.
At the AACR meeting, he said that the group has been working since then to further validate the method, which they call DELFI (DNA evaluation of fragments for early interception).
Among other work, Velculescu said the team is continuing to see good confirmation that the fragmentomic signals that it uses to identify cancer aren't confounded by non-cancerous mutations among white blood cells, or clonal hematopoiesis, potentially because these mutations don't affect DNA "packaging profiles" in the same way that a tumor does.
He noted that a challenge of comprehensive technical validation is that unlike with mutation methods where tumor tissue offers a source of "true positives" to benchmark against, there isn't the same reference readily available for something like a DNA fragmentation signal.
One thing that might work, Velculescu said, is to look at nucleosomal patterns from both white blood cells and tumor tissue in the same individual and try to see whether that reflects the cfDNA fragmentation patterns that are observed in the blood.
"But that's more difficult than it sounds," he said. Because the field doesn't as yet fully understand what nucleases are involved in digesting the genome to create cfDNA.
Luckily, he said, even if it's not possible to fully understand the biology and validate DELFI in that manner it should still be possible to validate and replicate its diagnostic performance, which the group is working on now.
Apart from the development of specific cancer detection assays, researchers also discussed emerging methods to improve epigenetic signal detection. Daniel De Carvalho discussed a method he and collaborators at Toronto's Princess Margaret Cancer Center developed called cfMeDIP-seq (cell-free methylated DNA immunoprecipitation and high-throughput sequencing), which they believe can help increase the accuracy and decrease the cost of methylation-specific sequencing tests.
Optimized from an existing low-input MeDIP-seq protocol to work with cell-free DNA, the approach involves enriching a sample for CpG-rich DNA fragments to avoid squandering sequencing reads unnecessarily across the whole genome.
In their initial work, De Carvalho and colleagues applied the approach to create a methylation classifier that could detect pancreatic cancer, lung cancer, and leukemia. But at the AACR meeting, he said more recent work has focused on brain cancer diagnosis, which although not necessarily an early-detection application, is also a challenging area for liquid biopsy because of biological factors like the blood-brain barrier.
In a separate poster at the meeting, investigators from Detroit's Henry Ford Health System provided their own look at how CpG methylation signals might provide a tool for glioma detection/diagnosis. The group developed a score to distinguish patients with glioma from those without and then applied it to an independent validation set of 68 blood samples from glioma patients, and 65 from non-gliomas, showing 97 percent sensitivity and 96 percent specificity.
On the commercial side, the AACR meeting also featured Grail's Eric Fung, who reviewed the company's most recent achievements in assay sensitivity from the development and validation study it calls the CCGA (circulating cancer genome atlas). After an exhaustive discovery study that analyzed the potential of both DNA mutation and copy number signals in addition to methylation, Grail ended up finalizing a targeted methylation classifier that uses bisulfite sequencing to analyze what it has determined to be the most informative regions of the genome.
In his presentation, Fung said that it has also been heartening to see that the finalized methylation assay derived from earlier CCGA discovery work performed similarly in a validation set of samples as it did in the training set reported last year at ASCO. Updated numbers for the assay performance in a pre-specified group of the most aggressive cancer types put it at 76 percent overall, with an average of 39 percent detection at stage I, 69 percent at stage II, 83 percent at stage III, and 92 percent at stage IV.
He also said the firm has been making progress on the tumor localization capabilities it first presented at last year's American Society of Clinical Oncology, specifically for pancreatic and hepatobiliary tumors for which the algorithm initially performed more poorly.
At the AACR meeting, Fung argued that Grail's pan-cancer approach is forward looking, in that it potentially helps avoid a future in which multiple tumor-specific screening assays layered on top of each other might combine to produce untenable false-positive rates, excess healthcare spending, and overdiagnosis.
But many of the other predominant firms in the space are focusing on single cancer types, a reflection of the boon that this strategy offers to test sensitivity.
Freenome, for example, also shared new data last month at the ASCO Gastrointestinal Cancers Symposium. In a poster, authors reported on results from a study of a subset of subjects enrolled in the firm's multi-center prospective study AI-EMERGE, which has included average-risk screening and case-control cohorts.
In the new analysis, the company analyzed samples from 43 subjects with CRC (11 of which were stage I and six of which were stage II) and 548 colonoscopy-confirmed negative controls using whole-genome sequencing, bisulfite sequencing, and protein quantification methods.
Among the 17 early-stage samples 16 were correctly classified, representing a sensitivity and specificity of 94 percent. Among the late-stage samples there was also only a single misclassification.
Subjects enrolled in the study were asked to provide both a blood and stool sample for a head-to-head comparison, and this allowed the Freenome researchers to also study their blood test against fecal immunochemical analysis.
Importantly, the authors wrote, only about half of the study participants even provided the requested stool sample for analysis, reinforcing prior observations of patient unwillingness to undergo stool-based testing. In those subjects with paired blood and stool samples, FIT achieved only about 67 percent sensitivity and 96 percent specificity compared to the much higher results for the firm's multi-omic blood-based approach.
Freenome has not been alone in eschewing a pan-cancer model for colorectal cancer, with Guardant Health staking its own claim, also with a multi-omic approach that incorporates methylation and fragment-based signals alongside cancer-associated DNA alterations. Meanwhile, Exact Sciences is exploring liver cancer detection via methylation and protein markers, and the Laboratory for Advanced Medicine is advancing a methylation-based classifier called IvyGene, also aimed at liver cancer.
The IvyGene team released some of its own new data late last year at a meeting hosted by the American Association for the Study of Liver Diseases.
Investigators tested blood samples from 450 subjects: 249 hepatocellular carcinoma patients distributed across stages, 83 healthy subjects, and 118 subjects diagnosed with benign cysts and nodules that could potentially produce confounding cfDNA methylation signals.
The test demonstrated an overall sensitivity of 88 and a combined specificity of 97 percent in distinguishing hepatocellular carcinoma samples from both healthy controls and patients with benign liver disease.
Sensitivity didn't appear to suffer for earlier-stage cancers, though these made up a relatively small proportion of the cohort. For Stage I, sensitivity was 81 percent, and for stage II it was 94 percent.
For all the commercial firms hoping to establish epigenetic or combined methods in early detection and screening, the next step will be to collect and share data from prospective studies that demonstrate performance in an intended use population.
Freenome has not yet started its planned registrational trial, but said last year that its intention is to conduct a prospective study benchmarking its test against colonoscopy.
In an email this week, Girish Putcha, the company's chief medical officer and clinical lab directly, said that the trial will begin this year and will enroll up to 14,000 patients from the intended use population in the US, aligned with America Cancer Society's recent recommendation to initiate colorectal cancer screening at 45.
Guardant, meanwhile, initiated its own 10,000-patient ECLIPSE (Evaluation of ctDNA LUNAR Assay In an Average Patient Screening Encounter) trial last October. And IvyGene has a 1,600-patient prospective study in the works comparing the sensitivity and specificity of its test alone, or in combination with ultrasound in a population at high risk of HCC due to liver cirrhosis.
Grail has had two prospective studies taking place for some time now: one in the US in women undergoing mammography and another in the UK split between those with no known cancer risk and those with elevated risk of lung cancer.
In addition to those two, the firm said last year that it would soon start returning results to physicians and patients within a third cohort. At the AACR meeting Fung disclosed the name of this effort: PATHFINDER.
According to the trial's online listing, which was posted on Jan. 27, it has already begun recruiting a planned 6,200 participants through two clinical partners: California's Sutter Health and Utah's Intermountain Healthcare.
About a third of the total cohort will be individuals 50 and older with no known elevated cancer risk. The rest will have a known elevated risk of cancer, including a history of smoking at least 100 cigarettes in their lifetime, documented genetic cancer predisposition, or a personal history of invasive or hematologic malignancy, with definitive treatment completed at least three years prior.
Physicians will order Grail's test and the company will return results to a study investigator. In cases with a "signal detected" test result the ordering and treating medical team will determine a diagnostic work-up based on the participant's clinical condition, recommendations by each institution's clinical practices, and in consultation with the study investigator and interdisciplinary care team.
Grail will follow participants for 12 months from the time of enrollment, assessing them again at that timepoint, and will track the number and types of diagnostic procedures required to achieve some sort of diagnostic resolution for any participants that get a positive result and subsequent follow-up. They'll also record participant-reported outcomes and perceptions of the experience.