Max Diehn, MD, PhD
During the 2021 International Society of Liquid Biopsy (ISLB) Congress, Max Diehn, MD, PhD, chief of radiation and cancer biology in the Department of Radiation Oncology at Stanford University, discussed whether the future of liquid biopsy included use as a tool for lung cancer screening.
Lung cancer remains the leading cause of cancer deaths worldwide, with fewer than 40% of patients diagnosed with early-stage disease. Instead, most are diagnosed with advanced stage disease, and survival outcomes in these patients are very poor, Dr. Diehn said.
“This suggests that if we could identify more patients at early stages, where outcomes are significantly better, we could likely save a significant number of lives,” he said. “For this reason, lung cancer screening has been heavily investigated for many decades, but it wasn’t until recently that we had a breakthrough with low-dose CT [LDCT] screening.”
LDCT vs. Liquid Biopsy
Studies have shown that LDCT has a sensitivity of about 95% and a specificity of about 85%. Dr. Diehn called this performance quite robust but added that uptake of LDCT screening is very low worldwide.
Reasons for this low uptake are varied, but key factors identified include lack of patient and clinician awareness, clinician concerns about LDCT radiation exposure, false-positive results, over-diagnosis, and cost.
Use of liquid biopsy could possibly overcome a number of these concerns, Dr. Diehn said.
“If we are going to try to implement liquid biopsy for lung cancer screening, how can that best be done in light of the existing effective screening method in the form of LDCT?” Dr. Diehn asked.
To succeed as a replacement, liquid biopsy would have to yield higher sensitivity and specificity than LDCT. If liquid biopsy was combined with LDCT, there are two approaches. First, LDCT could be used as a primary screening tool followed by liquid biopsy. In this case, liquid biopsy would have to have a sensitivity greater than or equal to LDCT and a specificity greater than LDCT. If liquid biopsy were used as a primary screening method followed by LDCT, then liquid biopsy would have to have a sensitivity greater than or equal to LDCT, but the specificity could be less than LDCT.
ctDNA
Next, Dr. Diehn focused on ctDNA as a liquid biopsy analyte.
“Cancer patients will often have low concentrations of tumor-derived DNA fragments in their circulation, and these can be noninvasively sampled with blood draws and detected using a variety of methods,” Dr. Diehn said.
The most prominent approaches developed to detect ctDNA include detection of somatic alterations, use of DNA methylation, and fragmentomics. Additionally, there are two main ways to perform ctDNA analysis—one that uses a tumor-genotype–naive approach, which requires no access to the tumor and would be used for screening, and a tumor-genotype–informed approach.
Dr. Diehn and colleagues have been working to develop a ctDNA-based screening approach. First, they studied how frequently ctDNA was present in patients with early-stage lung cancer, using a tumor-informed strategy. This revealed that most localized NSCLCs shed ctDNA, but levels are low, indicating the need for a very sensitive approach.
Next, they analyzed determinants of ctDNA shedding. They found that ctDNA concentrations were correlated with total body tumor burden as well as histology, with a higher shedding rate in non-adenocarcinomas compared with adenocarcinomas.
When they looked more specifically at adenocarcinomas, they found that ctDNA shedding was correlated with tumors that had a more solid radiologic appearance compared to those with a ground glass appearance.
Finally, additional analyses revealed that patients with detectable ctDNA had poorer outcomes overall compared with patients with low or undetectable ctDNA. This was true for stage I-III disease in aggregate and stage I disease alone.
“Detectable ctDNA pretreatment may be a useful biomarker for selecting patients for neoadjuvant therapy since [patients] with detectable ctDNA are highly enriched for having micrometastatic disease,” Dr. Diehn said.
Machine-Learning Model
Dr. Diehn and colleagues sought to identify key features to use in a machine-learning model to develop an early detection assay. To do this, they conducted a tumor-naive analysis of 74 control participants and 96 patients with early-stage NSCLC.
“In most patients with cancer, we found multiple mutations in the plasma, but strikingly, even in control participants, the majority also had mutations,” Dr. Diehn said. “Clearly just detecting mutations in plasma is not enough to say a patient is likely to have lung cancer.”
Further analysis revealed that the majority of mutations in the control participants were also present in the matched leukocytes of the blood sample. However, in patients with cancer, although there were a lot of mutations in the leukocytes, close to half of the mutations were present only in plasma, not white cells. This meant that most of the mutations in plasma came from clonal hematopoiesis, which needs to be taken into consideration when designing early detection studies using ctDNA.
Additional experiments revealed that mutational signatures and ctDNA fragment length helped distinguish lung cancer mutations from clonal hematopoiesis mutations.
Dr. Diehn and colleagues identified 20 features to combine in the machine-learning model and called it Lung-CLiP (Cancer Likelihood in Panel). They then trained the model with a cohort of 104 patients and 56 risk-matched control participants who had a negative LDCT screening. In this training cohort, at a 95% specificity, Lung-CLiP had a sensitivity for detection of lung cancer of about 40% for stage I, 55% for stage II, and 67% for stage III.
The researchers found the performance of Lung-CLiP competitive with a tumor-informed approach.
“The majority of patients whose tumors shed ctDNA were already being detected with the Lung-CLiP approach,” Dr. Diehn said.
An independent validation study found very similar performance and sensitivity for Lung-CLiP.
Conclusions
Based on these data, substituting liquid biopsy for LDCT will not be a valid screening approach, Dr. Diehn said. Neither will integration of the two approaches. Instead, he suggested the possible use of liquid biopsy in patients who refuse or are unable to get LDCT.
“If [liquid biopsy] is positive, then they can get LDCT,” Dr. Diehn said. “This might be the most promising approach.”
To conclude his presentation, Dr. Diehn emphasized that detecting more early stage cancer does not necessarily guarantee improved survival. Randomized studies are the only way to prove a survival benefit for ctDNA-based screening.
Dr. Diehn reported being a consultant for AstraZeneca, BioNTech, Boehringer Ingelheim, Genentech, Gritstone Oncology, Illumina, Novartis, RefleXion, and Roche. He is a founder and stockholder in CiberMed and Foresight Diagnostics, and has received grants or research support from AstraZeneca, Illumina, and Varian Medical Systems.
