NEW YORK – South Korean biotech Lunit published proof-of-concept research this month showing that its artificial intelligence platform could predict checkpoint inhibitor response based on the analysis of tumor infiltrating lymphocytes (TILs) in tumor tissue.
Researchers from Lunit and universities in Korea and China retrospectively evaluated the test, called Lunit Scope IO, in more than 500 non-small cell lung cancer patients and published their findings in the Journal of Clinical Oncology earlier this month. They focused specifically on patients with PD-L1 expression between 1 percent and 49 percent, considering that patients with greater than 50 percent of tumor cells expressing PD-L1 are already known to have improved responses to checkpoint inhibitors.
Lunit Scope IO analyzes the distribution of TILs in the tumor and tumor microenvironment. In the proof-of-concept study, researchers led by three senior authors — Tony Mok from the Chinese University of Hong Kong, and Yoon-La Choi and Se-Hoon Lee from Sungkyunkwan University in Seoul — sought to show that the AI-based platform can quickly analyze the spatial distribution of TILs in whole-slide images of tumor tissue, and by comparing them to thresholds of TILs in the cancer epithelium and stroma, segment patients into three phenotype groups.
In the inflamed phenotype group, TILs are distributed intratumorally; in patients with the immune-excluded phenotype, TILs are excluded from within the tumor by stromal cells; and in those with the immune-desert phenotype, TILs are scant in the tumor microenvironment.
The researchers trained the algorithm underlying the Lunit test on more than 3,100 whole-slide images of 25 cancer types, including lung cancer samples from The Cancer Genome Atlas, Sungkyunkwan University's Samsung Medical Center, and Seoul National University's Bundang Hospital.
To determine if the algorithm could predict checkpoint inhibitor response, the researchers retrospectively analyzed 518 patients with advanced NSCLC who received checkpoint inhibitor treatment at Samsung Medical Center and Seoul National University Bundang Hospital. Patients received a variety of drugs in different treatment lines, though most received either Merck's Keytruda (pembrolizumab), Bristol Myers Squibb's Opdivo (nivolumab), or Genentech's Tecentriq (atezolizumab). Most patients also received checkpoint inhibitors in the second- or third-line setting.
The analysis yielded a 26.8 percent overall response rate to checkpoint inhibitors in patients with the inflamed phenotype. Comparatively, 11.5 percent of immune-excluded patients responded to checkpoint inhibitors, and 11.2 percent of immune-desert patients responded to such drugs.
Median progression-free survival in the inflamed group was 4.1 months and median overall survival was 24.8 months — approximately twice as long compared to these outcomes in the other two groups.
However, in this initial analysis, the researchers noted that the inflamed phenotype had significant overlap with patients who had PD-L1 expression greater than 50 percent. They further explored the data to determine whether patients with lower PD-L1 expression but an inflamed phenotype would also derive a greater benefit from checkpoint inhibitors.
They found that 42.5 percent of patients with PD-L1 expression between 1 percent and 49 percent were classified as having an inflamed phenotype. In that group, the response rate was 22.8 percent compared to a response rate of 3.9 percent in patients with the same PD-L1 expression level but a noninflamed phenotype.
"We think AI analysis of TILs could complement PD-L1 IHC testing," Lee, one of the senior authors of the study and a professor at the Sungkyunkwan University School of Medicine, said in an email. "We can dissect the PD-L1 tumor proportion score between 1 percent and 49 percent group in a more detailed manner."
The researchers concluded in the paper that the AI-enabled TIL-based phenotypes could serve as a "relatively low-cost and efficient biomarker" to analyze hematoxylin and eosin-stained slides for NSCLC patients.
In practice, Lee suggested that Lunit Scope IO could be used with PD-L1 immunohistochemistry testing, which is already a standard part of the diagnostic workup for advanced NSCLC patients for guiding therapy. If the hematoxylin and eosin-stained slides are adequate for analysis, he noted that the Lunit Scope IO results can be turned around within one day.
"The present lung cancer guidelines recommend PD-L1 [testing] along with [assessment of] driver mutations," Lee continued, adding the results of the present study suggests that the "Lunit Scope IO can be used along with PD-L1 to select initial treatments."
Although PD-L1 expression has become a standard test performed in lung cancer patients, many oncologists don't feel it is a reliable biomarker for predicting which patients will respond to checkpoint inhibitors.
A 2017 review of predictive biomarkers for checkpoint inhibitor response suggested that PD-L1 expression measured by IHC has several limitations. For example, there are a wide range of thresholds depending on the drug and tumor types, from greater than 1 percent to greater than 50 percent, that physicians need to consider. The review also pointed out that PD-L1 expression is regulated by various mechanisms, that it can be heterogeneous within a single tumor, and that IHC does not take into account immune markers or tumor microenvironment factors that could also affect checkpoint inhibitor response.
However, the US Food and Drug Administration has approved several drug indications based on different PD-L1 expression levels. In advanced NSCLC, for example, Keytruda and Opdivo are both approved for patients with PD-L1 expression greater than 1 percent, while Tecentriq was approved for PD-L1 expression greater than 50 percent.
These issues related to PD-L1 has oncologists eager for biomarkers that can help them determine more definitively which NSCLC patients will derive benefit from immunotherapy treatments.
Lunit Scope IO is currently only available for research use, but the company will conduct additional trials to validate the findings of this proof-of-concept study and expand use of the test to other cancer types, said Chan-Young Ock, Lunit chief medical officer and one of the authors of the published study, in an email.
"We plan to expand the range of our research to validate the efficacy of Lunit Scope IO in all cancer types originated from epithelium, such as breast cancer, colorectal cancer, liver cancer, stomach cancer, and others," Ock said. "We are aiming to obtain [regulatory] approval for Lunit Scope IO by 2024 and will prepare for commercialization immediately afterward."