NEW YORK – A novel ctDNA detection assay developed by researchers at Stanford University could help physicians screen individuals at risk for lung cancer.
Described in Nature this week, the method, called Lung-CLiP (lung cancer likelihood in plasma), involves targeted sequencing of cell-free DNA from plasma and matched white blood cell DNA to assess copy number and single nucleotide variants, coupled with a machine learning model that estimates the probability that a cfDNA mutation is tumor-derived. The estimate is based on biological and technical features specific to each variant, such as background frequency, cfDNA fragment size, the gene affected, and the likelihood of clonal hematopoiesis.
The results of the SNV model and the genome-wide copy number calls are then integrated to yield a final "Lung-CLiP score."
Notably, the researchers have found that so far, their method could detect the presence of tumor DNA in the blood at a similar rate as methods that generate tumor tissue sequencing data first to develop targeted panels for blood-based mutation detection. Those other methods are already entering into the clinic in adjuvant and cancer monitoring settings.
Current recommendations for lung cancer screening include the use of annual CT scans for individuals known to have an elevated risk based on factors like a history of smoking. But while this imaging surveillance has been shown to reduce lung cancer deaths, uptake remains relatively low.
According to the authors of the new Stanford study, only around 5 percent of eligible individuals undergo radiographic screening in the United States due to factors such as high costs, limited access, and concerns with false positives.
Genomic blood tests for cancer screening and early detection have become the focus of attention in the molecular diagnostic space, though most activity so far has been either toward pan-cancer screening tools, or in a few other specific tumor types like colorectal cancer, where tests hope to vie against colonoscopy and existing stool-based methods.
In their new study, the Stanford investigators, led jointly by Maximilian Diehn and Ash Alizadeh, have now proposed a method that harnesses some of the same principles for lung cancer, sharing preliminary but promising validation results.
The team first trained and optimized Lung-CLiP in an initial sample of 104 patients with early stage non-small cell lung cancer and 56 matched controls.
When they then applied it to an independent set of validation samples (46 cases and 48 risk-matched controls), the test was able to discriminate early-stage lung cancer patients with sensitivity and specificity levels that the authors believe suggest a significant benefit to the clinic: depending on where they set their specificity threshold, the method could achieve 63 percent sensitivity for stage I tumors and up to 75 percent sensitivity in detecting patients with stage III disease.
Because the LungCLiP test is algorithmic, it can be tuned in one direction or the other to match different specificity cutoff points, and the authors reported sensitivity calculations using two of these that they picked to reflect two possible clinical implementations: a standalone screening scenario in which "high specificity would be desirable to minimize false positives," and a screening scheme where "a lower specificity may be acceptable if Lung-CLiP were applied to the approximately 95 percent of at-risk individuals who are not currently undergoing [CT] screening owing to access limitations or other obstacles," the authors wrote. Patients with positive Lung-CLiP tests would then be referred for follow-up imaging.
Setting their cutoff at 98 percent specificity to reflect the first scenario, the researchers found that Lung-CLiP detected 41 percent of patients with stage I disease, 54 percent of patients with stage II disease and 67 percent of those with stage III disease.
This performance exceeds at least some of the initial results that have been described for lung cancer using other methods now being advanced for cancer screening.
Early detection firm Grail, for example, while its goal is to create a pan-cancer screening product, has shared some breakout data specific to lung cancer from its ongoing Circulating Cell-free Genome Atlas study in recent years.
In a 2018 presentation, for example, the company pulled out a case-control cohort of 561 non-cancer subjects and 118 patients with lung cancer. According to that study, detection at a 98 percent specificity cutoff only reached about 51 percent for participants with early-stage lung cancers (stages I-IIIA).
In their study this week, the Lung-CLiP investigators also analyzed sensitivity at 80 percent specificity, representing their second proposition of a CT-reinforced screening model. At this setpoint, they observed even better sensitivities: 63 percent in patients with stage I disease, 69 percent in patients with stage II disease, and 75 percent in patients with stage III disease.
In an email, Alizadeh said that while Lung-CLiP, as it stands, may not undercut the cost of CT scans, its relative ease positions it to help fill the significant gap left by radiographic screening wherein the vast majority of individuals who are eligible still fail to receive it.
"Lung-CLiP [could] help increase the rate of early detection," he wrote. "This would be analogous to how stool-based testing [proposes to] improve screening for colorectal cancers, especially in populations where adoption of colonoscopy is lower than currently recommended."
This type of hybrid approach "could potentially increase the total number of patients screened and therefore the number of lives saved annually in the United States, from the current annual value of around 600 to closer to the projected maximum of around 12,000," the study authors wrote.
That said, many steps remain for the team in seeing the clinical potential of Lung-CLiP realized.
According to Alizadeh, for a case to be made to regulatory and guideline bodies like the US Preventive Services Taskforce, for example, the group will need to validate the approach prospectively.
"We are working with our collaborators on the design and execution of a prospective validation study in a larger cohort of adults at risk for lung cancer," he said.
Apart from the method's detection rate, another notable study result was the fact that although ctDNA levels were very low in early-stage lung cancers, it was present prior to treatment in almost all the confirmed cancer cases in the cohort. In other words, a complete lack of circulating tumor DNA did not appear, in this group at least, to be the issue preventing early cancer detection.
According to Alizadeh, this means that if multi-analyte, machine learning inference methods like Lung-CLiP can be pushed successfully toward lower and lower limits of detection, they could hopefully achieve reliable detection for a majority of cases, even at these very low ctDNA levels.
Although the Lung-CLiP method involves the analysis of DNA fragment size, it doesn't incorporate methylation, which many others in the field have embraced to try to push sensitivity beyond what is possible with somatic variant detection alone.
According to Alizadeh, this was because most available methylation-based assays require a different molecular biology workflow and a second separate sample of cfDNA, which the team's current study couldn't support.
However, he said, "this is an active area of research by us and others. We think that it is possible that the broader integration of mutations, copy number alterations, fragmentomics, and methylation within a unified assay could further improve detection performance."