This story has been updated from a previous version for content and clarity.
NEW YORK – A group of European researchers led by the Cancer Research UK Cambridge Institute has developed an analytical method to improve the detection of circulating tumor DNA (ctDNA) by tracking large numbers of patient-specific cancer mutations, a method which the team believes could be used for minimal residual disease (MRD) and cancer recurrence testing.
The method, called "INVAR" (INtegration of VAriant Reads), applies both error-suppression methods and signal-enrichment approaches to analyze hundreds of patient-specific mutated loci to detect ctDNA in plasma samples.
The group published a proof-of-principle study last week in Science Translational Medicine that highlighted INVAR's ability to increase detection of mutations in cancer patients. In addition, the team presented data yesterday at the virtual 2020 American Association of Cancer Research annual meeting suggesting the method could detect ctDNA in non-small-cell lung cancer (NSCLC) patients prior to curative treatment.
While the concept of targeted, patient-specific mutation panels for MRD and cancer recurrence testing isn't exactly new, INVAR can dramatically increase sequencing sensitivity by tracking several hundred, if not thousands, of tumor mutations per patient, noted Nitzan Rosenfeld, study senior author and group leader at the Cancer Research UK Cambridge Institute.
After collecting DNA from tumor biopsy and matched normal (buffy coat), and 2 to 3 ml of a blood sample from a patient per timepoint, Rosenfeld's team performed a combination of whole-exome, whole-genome, and targeted panel sequencing on the samples using a combination of Illumina NovaSeq and HiSeq instruments.
While other techniques apply machine learning, INVAR instead uses a statistical approach to analyze the sequencing data, explained Jonathan Wan, study first author and recent University of Cambridge MD/PhD graduate.
INVAR combines error-suppression tools, such as read collapsing and unique molecular identifiers, to reduce errors in patient-specific sequencing data. The method measured ctDNA "integrated mutant allele fractions" (IMAF) by taking a background-subtracted, depth-weighted mean allele fraction across patient-specific tumor-mutated loci in each sample.
The method compares the mutations found in a patient's tumor and blood samples. INVAR then boosts ctDNA signal by assigning greater weight to loci with higher mutated allele fractions in tumor sequencing data, and sequencing reads in plasma that most closely match the size distribution data in ctDNA.
As proof of principle, they used INVAR to track on average around 700 mutations in each patient's blood sample, but they demonstrated that the method can also be used to monitor upwards of 5,000 mutations per patient.
"By knowing a patient's mutational profile in advance, we know what to look for in their bloodstream in a targeted manner," Wan explained. "INVAR takes these things together, integrates variant reads (hence the name), then gives you a yes or no classification [for cancer detection]."
Katrin Heider, study co-author and post doctorate research associate at the University of Cambridge, acknowledged that the team encountered issues with minimizing background noise while developing INVAR.
"You have to control the noise while making sure that you're not drowning out or removing the true signal you might have," Heider said. "We're trying to strike a balance between filtering out reads that are most likely noisy while maintaining the reads of interest."
Rosenfeld's team therefore used statistical information derived from INVAR to determine the strengths of informative reads. Based on the data, the researchers could observe that mutations with higher allele frequency in tumors were more likely to be seen in plasma. If the mutation was spotted in the plasma and the researchers saw that the mutation had a high allele frequency in the tumor data, they could be more confident that it was tumor-derived.
Between the wet lab and computational analysis portions, Wan estimated that once panels have been generated the overall INVAR process could take about a week to process samples and check for presence of ctDNA.
In the study published last week, Rosenfeld's team applied INVAR to sequencing data from 176 plasma samples from 105 cancer patients, as well as 45 healthy control samples. Cancer types included melanoma, breast, renal, NSCLC, and gliomas.
Collecting at least one tumor biopsy and plasma sample from each patient, the researchers first performed tumor tissue genotyping to identify patient-specific mutations. They then used the mutation lists to create custom capture panels for sequencing longitudinal plasma samples from each patient.
By integrating signals across a median of greater than 100,000 informative reads targeting tumor-identified mutations, the team successfully quantified ctDNA to one mutant molecule per 100,000 molecules. In cases with high tumor mutation burden or high plasma input material, the team even quantified ctDNA to individual parts per 1 million molecules.
Rosenfeld's team applied INVAR to custom capture panel sequencing data in plasma samples from patients with stage II to stage IV melanoma to detect ctDNA, in some cases trace amounts of it.
Overall, INVAR produced a median area under the curve (AUC) of 0.80 for early-stage cancers and an AUC of 0.98 for late-stage cancer.
The team then tested INVAR for personalized monitoring in melanoma patients in response to first-line anti-BRAF targeted therapy. In some patients, the researchers identified differential responses of mutation clusters to targeted therapy, indicating a varied response to the therapy in different tumor subclones.
To explore the potential use of INVAR for MRD testing, Rosenfeld's team applied the method to samples from patients at high risk of recurrence after resection of stage II to III melanoma. Of the 35 evaluable patients, the team detected ctDNA in eight of 20 who later recurred but did not show a significant difference in disease-free intervals. Because of the small sample size and limited power of the statistical analysis, the study authors noted that validation studies will be needed to benchmark the approach in larger melanoma cohorts and in other cancer types after surgery.
Rosenfeld's team then aimed to see if INVAR could be integrated as part of WES and WGS workflows for personalized ctDNA monitoring.
The researchers first selected samples with IMAF values between 0.000045 and 0.16 based on custom capture sequencing, and sequenced them by whole-exome capture.
The group detected ctDNA in all 21 tested samples with IMAFs as low as 0.000043 with a specificity of greater than 95 percent, suggesting that INVAR could identify ctDNA from WES data using patient-specific mutation lists.
Rosenfeld's team then performed low-depth WGS (0.6x mean depth) on libraries prepared from cell-free DNA of longitudinal plasma samples from six patients with stage IV melanoma, who had more than 500 patient-specific mutations identified using tumor WES. Using INVAR on the WGS data, the team quantified IMAF values as low as 0.0011 with a threshold specificity of 97 percent.
However, Heider noted that INVAR is not ideal to screen early-stage cancers because the technique requires an existing list of patient-specific mutations. Because INVAR has only been applied to a limited number of cases, the team also believes it might have contributed to the limited statistical power for measuring MRD in early-stage melanoma in the study.
"Patient-specific mutation lists provide an opportunity for highly sensitive monitoring from a range of sequencing data types using methods for signal aggregation, weighting, and error suppression," the study authors said. Such lists may also be increasingly leveraged for individualized monitoring from a variety of sequencing data types for sensitive monitoring, they noted.
Viktor Adalsteinsson, associate director of the Broad Institute's Gerstner Center for Cancer Diagnostics who was not involved in the study, believes that Rosenfeld's team demonstrated an "elegant method" that adds to the growing body of evidence that mutations can be highly specific tumor markers and that tracking more mutations per patient in cell-free DNA can improve sensitivity.
"Being able to track cancer burden from a liquid biopsy could have a profound impact on patient care," Adalsteinsson said in an email. "[Rosenfeld's team] also describe strategies to control sequencing error rates, which will be crucial both as lower tumor fractions are sought to be detected and as the field looks to use ultrasensitive liquid biopsy tests to guide patient care."
Dan Landau, an assistant professor at Weill Cornell Medicine and core faculty member at the New York Genome Center who was not involved in the study, noted that low input material typically poses a formidable challenge to ctDNA detection using targeted panels that are advised by a small amount of informative sites. He therefore believes that INVAR represents advances in the sequencing field that will inspire current patient-specific methods to widen their scope, allowing improved sensitivity in residual monitoring.
"[INVAR] shows that the increase of targeted panels to hundreds of targeted sites can transform detection of ctDNA by overcoming the ceiling imposed by the limited input of depth of sequencing for each site," Landau said in an email. "[The method] offers a suite of de-noising tools to ensure that the expansion to a large number of targeted sites will not drown the added signal with sequencing-related noise."
However, Landau argued that an issue with INVAR for potential widespread clinical applications may involve the "technically challenging" creation of large patient-specific panels. While using INVAR may increase the turnaround time at critical therapeutic windows, such as the curative window for post-operative adjuvant therapy, he believes that the method's need to validate patient-specific performance may lead to bottlenecks for clinical applications.
AACR LUCID study
At the 2020 virtual AACR conference, Davina Gale, senior research associate in Rosenfeld's team, presented the results of applying INVAR to detect ctDNA in early-stage NSCLC. The group aimed to evaluate ctDNA distribution in baseline plasma samples from NSCLC patients before treatment.
In the Lung cancer-Circulating tumor DNA (LUCID) study, co-led by Robert Rintoul, a respiratory physician at the Royal Papworth Hospital, Rosenfeld and his colleagues recruited 100 patients with stages I through III NSCLC, collecting plasma samples prior to and after cancer therapy. After sequencing the samples using patient-specific hybrid capture assays, the team analyzed the data using INVAR. To assess ctDNA in patients without available tumors, the group also analyzed plasma samples with Inivata's inVisionFirst-Lung assay.
Using both platforms, Rosenfeld's team identified ctDNA signals in 66 of the 100 patients prior to treatment, with ctDNA levels as low as nine parts per million with >95 percent specificity. The group detected ctDNA in 52 percent of 60 patients with stage I NSCLC, 86 percent of patients with stage II cancer, and 90 percent of stage III patients. By analyzing different histological subtypes, the team also detected ctDNA in 79 percent of squamous cell carcinomas and 60 percent of adenocarcinomas.
Rosenfeld's team believe that INVAR may be able to detect ctDNA in most patients with early-stage NSCLC prior to treatment. Rosenfeld and his colleagues have also tested INVAR on other samples from cancer patients, including glioblastoma and breast cancer.
In terms of comparing INVAR to existing techniques, Rosenfeld believes their INVAR results compare favorably to data recently presented using other patient-specific sequencing analysis methods such as the one developed by Adalsteinsson's team at the Broad Institute and the Dana-Farber Cancer Institute, as well as MRDetect developed by Landau's team at the NYGC and Weill Cornell. Investigators from TGen, led by Muhammed Murtaza, co-director of TGen's Center for Noninvasive Diagnostics, have also developed a liquid biopsy technique called TARDIS (targeted digital sequencing) to monitor patietns with early-stage cancer.
Inivata, where Rosenfeld serves as chief scientific officer, is developing an amplicon-based sequencing tool called RaDaR, which can track up to 48 patient-specific mutations. In a separate poster presented by Dr Gale at AACR, the LUCID team used the patient-specific RaDaR assay to detect MRD after treatment of NSCLC patients. Meanwhile Natera has also spearheaded MRD and cancer recurrence detection efforts with its Signatera cell-free ctDNA assays, which also integrate the creation of individualized assays based on patient tumor tissue exomes to boost ctDNA detection sensitivity.
Highlighting that INVAR is cancer- and mutation-agnostic, Rosenfeld believes the method could be used in the clinical space for a variety of different cancers. While the researchers have filed a patent for INVAR, they have not yet decided which specific commercialization option to pursue.
"We'll be looking to see how such sensitive methods [such as RaDaR and MRDetect] progress and what results these yield, where there are gaps [that] could benefit from further improvement to sensitivity, or areas where INVAR could have a specific advantage due to the biology or clinical path of the disease," Rosenfeld said. "In the meantime, we are focusing on applying this method as a research tool to study ctDNA in early and pre-symptomatic disease."
Rosenfeld is also interested in combining INVAR and similar personalized ctDNA detection methods to develop cancer monitoring tests using a patient's pinprick blood sample. He believes an integrated method could potentially allow for an increase in routine cancer monitoring frequency, or could provide an option for patients who are unable to get to clinic for routine assessments.