NEW YORK – Using extracellular vesicles extracted from blood, researchers believe that they have found a way to improve the commonly used biomarker PD-L1 and better predict responses to cancer immunotherapy in patients with non-small cell lung cancer.
The data, published earlier this month in the Journal of Experimental and Clinical Cancer Research, showed that changes in the prevalence of PD-L1-expressing exosomes early on in immunotherapy treatment were a better predictor of progression-free survival than PD-L1 status as measured in tumor tissue samples. Adding radiomic analysis of pretreatment imaging scans even further improved the ability to classify responsive patients.
If validated in larger cohorts, the study authors believe their approach could supplant tissue PD-L1 as the standard of care for guiding immunotherapy use in NSCLC and potentially other tumor types.
PD-L1 remains a mainstay for assessing patient eligibility for immunotherapy, even with the emergence of more recent tools like tumor mutational burden testing. Despite this, the biomarker has long been regarded as woefully imperfect. "The problem with PD-L1 in tissue is that it's very heterogeneous," said Christian Rolfo, professor of medicine at Mount Sinai's Icahn School of Medicine and lead author of the new exosome study. In two samples from the same tumor, one could be strongly positive and the other nearly negative.
According to Rolfo, data from past trials have shown that among patients with more than 50 percent PD-L1 positivity in stained tissue sections (a commonly used cutoff to predict response) there are individuals who don't respond as well, or as quickly, or the same way. Similarly, patients with less than 50 percent positivity don't always lack response altogether. Specific cases are seen responding while others do not.
Often tissue samples used for PD-L1 staining are remnants from initial diagnostic biopsy, not reflecting changes in expression that might have occurred after first-line treatments. "A single tissue biopsy may not be able to recapitulate the exact status of the tumor microenvironment at the time of treatment that, in some cases, can be months or even years after tumor collection," the study authors wrote.
Rolfo and his colleagues wanted to explore whether blood could also supply a better platform for PD-L1 analysis, turning to extracellular vesicles, or exosomes, to provide a source for detection of the biomarker. According to the group, evidence from other research has suggested that EVs may play an important role in tumor progression and tumor‐immune interaction.
"You can do [circulating tumor cells], but they are a little more complicated to isolate. Exosomes seem to be much simpler with the technology we are using," Rolfo said.
Although other studies have explored PD-L1 in exosomes, none has previously demonstrated that the dynamics of EV PD-L1 could be a reliable predictive biomarker in patients with NSCLC and could potentially outperform the current standard tissue PD-L1, Rolfo and his colleagues wrote.
Other blood-based biomarkers have also been explored, including tumor mutational burden, but assay standardization has long been a problem for the biomarker, and some recent analyses have raised questions about broad utility.
In their study, Rolfo and his team collected blood samples at two time points — before treatment and at the ninth week of treatment — from two cohorts of patients with non-small cell lung cancer.
First, investigators analyzed 33 retrospective samples from patients treated with immune checkpoint inhibitors as a training cohort. They then added 39 more as a prospective validation group from the Phase II PROLUNG clinical trial at the National Cancer Institute, Mexico, including 25 patients treated with immunotherapy plus chemotherapy and 15 who received chemotherapy alone as controls.
After extracting EVs using ultracentrifugation, the researchers measured the level of PD-L1 protein expression using immunoblotting. They calculated EV PD-L1 dynamics by dividing a normalized measure of PD-L1 expression at the nine-week time point by the same value in patients' paired baseline samples.
Authors reported that EV PD-L1 increased during treatment in the first cohort's non-responders and decreased in its responders. The same was true among immunotherapy-treated patients from the PROLUNG trial, while no differences were seen in the chemo-only controls. Investigators calculated that across all 57 immunotherapy-treated patients PD-L1 dynamics had 73 percent sensitivity and 61 percent specificity.
Patients with an EV PD-L1 decrease had longer progression-free survival than those with increasing levels, with a hazard ratio of 0.36 in the training cohort and 0.18 in the PROLUNG patients. A multivariate analysis for the 57 total patients revealed that change in EV PD-L1 was an independent predictive marker of PFS. Similar results were seen for overall survival.
Looking at patients' tissue, meanwhile, there was no association between PD-L1 expression and either survival measure.
Having done previous studies showing that radiomics can also predict immunotherapy response, the group performed an exploratory analysis using baseline CT scans from 27 patients in the training cohort, reporting higher sensitivity using the two measures combined.
Among caveats, Rolfo recognized that the study was small so should be interpreted with caution right now. But he and his team are already working on a validation study in a separate cohort from an immunotherapy study where blood samples are available from several time points.
The group also intends to recruit a larger cohort for even further validation. "We are trying to get funding for that now, which is a challenge … but now that the paper is published, there may be some more opportunities there," Rolfo said.
The concept of an on-treatment biomarker could also be a challenge for clinical implementation among clinicians who are accustomed to employing predictive testing before patients begin treatment.
But Rolfo said the oncology community is already becoming comfortable with dynamic on-treatment liquid biopsy in other contexts, for example, using genomic tests to assess response or identify emerging recurrence.
He and his team also plan to publish another paper analyzing the predictive power of blood-based PD-L1 assessment at a single pretreatment time point, where he said they also saw good results, though potentially not as powerful as exosomal PD-L1 dynamics.
However, the authors acknowledged that isolation of EVs and analysis of protein expression by western blot would be complex to apply in a clinical setting. But this could be overcome with the future development of microfluidic devices allowing on-chip ELISA detection.
As many others in the immunotherapy field have expressed, Rolfo said that it's most likely the future will see more and more combinations of different technologies, reflecting the fact that no one biomarker or component has seemed to be able to account for the entirety of immunotherapy responsiveness.
"That's in part how we envisioned adding the radiomics to the model," he said, adding that the team is also now looking at whether RNA-seq of exosomes and measurement of ctDNA can add value by providing measures of mutational load and gene expression.