NEW YORK – Researchers at the Ohio State University Comprehensive Cancer Center – The James are seeking funding and a pharmaceutical partner for a prospective study to gauge how well cancer patients with structural variations in PD-L1 and PD-L2 respond to anti-PD-1 treatment.
Based on research conducted so far, a multi-institutional team led by OSU physician-scientist Sameek Roychowdhury has hypothesized that PD-L1/2 structural variants may be biomarkers that can identify which patients with solid tumors, leukemias, and lymphomas respond to anti-PD-1 checkpoint inhibitors. "We definitely think this is worth exploring as a predictive marker of good response to PD-1 immunotherapy in a pan-cancer way," said Roychowdhury, a member of OSUCCC's translational therapeutics program.
Although immunohistochemistry-based PD-L1 expression is commonly used to guide anti-PD-1 or -PD-L1 therapy for patients with solid tumors, many oncologists say it's not precise enough for distinguishing responders from non-responders and often bemoan the need for better predictive biomarkers. Meanwhile, microsatellite instability (MSI), mismatch repair deficiency, and tumor mutational burden of at least 10 mutations per megabase can be used in a histology-agnostic way to identify patients with refractory solid tumors eligible for certain checkpoint inhibitors, such as Merck's Keytruda (pembrolizumab).
When considering whether to prescribe anti-PD-1/PD-L1 drugs to patients with blood cancers, however, oncologists typically don't test for predictive biomarkers. Keytruda, for example, is approved for all refractory classical Hodgkin lymphoma and primary mediastinal large B-cell lymphoma patients. But as Roychowdhury and colleagues pointed out in their JCO Precision Oncology paper published on Monday, a significant subset of primary mediastinal large B-cell lymphomas have PD-L1/2 structural variants, and even though the US Food and Drug Administration approved Keytruda based on a 45 percent overall response rate in an all-comer population, the researchers suspect that these responses may be driven by patients with PD-L1/2 structural variants.
In this recently published analysis, Roychowdhury and colleagues identified around 400 cancer cases — 274 with solid tumors and 212 with lymphomas — with PD-L1/L2 structural variants. To identify these cases, they scoured 22 papers and searched sequencing data housed within The Cancer Genome Atlas (TCGA), the Oncology Research Information Exchange Network, the International Cancer Genome Consortium, and the tumor genomic profiling company Foundation Medicine (a coauthor on the JCO Precision Oncology paper). In these datasets, they observed PD-L1/2 structural variants most frequently in diffuse large B-cell, NK or T-cell, and primary mediastinal B-cell lymphomas, as well as in lung, breast, and head and neck tumors.
Based on their research so far, Roychowdhury's group estimates that between 0.5 percent and 1 percent of cancers may harbor PD-L1 structural variants. That means that these alterations may be important for informing anti-PD-1 treatment for between 90,000 and 180,000 cancer patients worldwide each year.
Moreover, Roychowdhury pointed out that the frequency with which these structural variants occur in cancers is on par with other precision oncology targets such as NTRK, RET, and MSI-high, and therefore warrant further evaluation as a predictive marker of immunotherapy response across cancer types. "The clinical impact of immunotherapy for this molecular subset has the potential to be similar to MSI-high [status associated with] amazing durable responses and is a diagnosis an oncologist never wants to miss," he said.
The PD-L1/2 structural variations Roychowdhury's group identified were diverse in size and structure and included duplications, deletions, translocations, and inversions, as well as some unclassified variations. They tended to occur in hotspots in introns 4, 5, and 6, though when it came to translocations, researchers didn't identify a pattern in terms of common gene fusion partners with PD-L1 and PD-L2 variants.
Past studies have suggested that patients with PD-L1 structural variants were likely to have PD-L1 expressing tumors. Roychowdhury's group confirmed this using TCGA data.
Using CIBERSORT, an in silico tool that estimates immune cell type proportions in a sample, his team further noted that PD-L1 and PD-L2-altered tumors were more likely than wild-type tumors to have inflamed immune microenvironments and be enriched for CD8-positive T cells, regulatory T cells, resting natural killer cells, activated CD4 T cells, and M1 macrophages.
The identification of PD-L2 structural variants and their biological impact was of particular interest to Roychowdhury's team in this study because the field has paid very little attention to this ligand. Patients considering immunotherapy are commonly tested for PD-L1 expression in tumor and immune cells using IHC but not for PD-L2 expression.
"Right now, we have a biased world view of PD-1 and PD-L1. We only look at ligand one," Roychowdhury said. "We may need to reconsider PD-L2 in both solid tumors and leukemias and lymphomas."
The two top checkpoint inhibitors on the market, Keytruda and Bristol Myers Squibb's Opdivo (nivolumab), are designed to block the interaction between PD-1 and both its ligands, PD-L1 and PD-L2. But there are drugs like Roche's Tecentriq (atezolizumab), AstraZeneca's Imfinzi (durvalumab), and EMD Serono/Pfizer's Bavencio (avelumab) that just inhibit the interaction between PD-1 and PD-L1. A patient whose tumor isn't overexpressing PD-L1 but is using PD-L2 to evade the immune system is unlikely to benefit from a PD-L1-specific treatment, Roychowdhury reflected.
Finally, Roychowdhury's team turned to the question of how patients with these variants did on immunotherapy. They manually curated a dozen studies and identified 71 patients who had received anti-PD-1 treatment and had solid tumors and lymphomas characterized by PD-L1/2 deletions, partial duplications, translocations, and inversions, as well as copy number amplifications. Among these patients, 52 patients had complete or partial responses that lasted for more than a year.
"More than half of these cases had really good responses that were very durable," Roychowdhury said.
Diagnostic, clinical trial implications
Roychowdhury's team is already thinking about the implications their initial findings may have on the immunotherapy trials underway and the manner in which patients are selected for treatment. Even though some of the first PD-L1/2 structural variants were identified in classical Hodgkin and mediastinal large B-cell lymphomas, and patients with these types of blood cancers tend to have the best responses to anti-PD-1 treatments, many past anti-PD-1 therapy trials in lymphoma patients haven't involved biomarker selection. But this latest analysis suggests that PD-L1 structural variants may be associated with exceptional response to anti-PD-1 treatment among lymphoma patients.
Even though it is currently not standard of care to test for structural variants in PD-1 ligands, Roychowdhury and colleagues in their paper proposed "that the prospective selection of patients with non-Hodgkin lymphoma on the basis of structural variations may identify a molecular subset of patients who can benefit from PD-1 immunotherapy."
Furthermore, even though patients with PD-L1/2 structural variants tend to have high PD-L1 expression greater than 50 percent, patients who have no or very low PD-L1 expression by IHC tests on the market may still harbor structural variants that make them good responders to immunotherapy. These routine PD-L1 IHC tests would also miss patients with PD-L2 structural variants.
"IHC is imperfect. You can be PD-L1 low for IHC and have an amazing response to anti-PD-1 treatment," Roychowdhury said. His team didn't compare whether PD-L1 IHC or sequencing for PD-L1/2 structural variants could better identify best responders to immunotherapy, but the analysis certainly shows that IHC-based PD-L1 expression testing alone is not enough. "IHC in conjunction with knowing structural variant or copy number status may give you a better idea [of best responders]," Roychowdhury said, noting that today it's not uncommon for patients to have PD-L1 IHC testing alongside comprehensive genomic profiling.
However, his team's analysis also points to the importance of ensuring that next-generation sequencing-based genomic profiling tests can detect these structural variants. "Since these alterations involve both exons and introns, one of our current projects is to optimize sequencing and bioinformatics methods to ensure we are detecting [them]," Roychowdhury said.
What Roychowdhury's team has right now, is "strong" retrospective evidence, he said, but what they need next is to prospectively study how well patients with PD-L1/2 structural variants do on anti-PD-1 therapy. The study his group has designed and is seeking pharma partners for would have two cohorts, one enrolling solid tumor patients with PD-L1/2 structural variants and another including solid tumor patients with PD-L1/2 amplifications. Patients in the study would receive an anti-PD-1 drug that could block both PD-L1 and PD-L2. They would get a CT scan every 12 weeks to track their responses to the treatment and have ctDNA testing to gauge minimal residual disease every three weeks.
"Instead of waiting a year to find out durability, we may have some early signs of response in ctDNA," Roychowdhury said.
It remains to be seen whether current sponsors of marketed checkpoint inhibitors are interested in exploring new predictive biomarkers, however. The top-selling immunotherapies have already achieved market approval in multiple cancer types using PD-L1 expression, however imperfect, as well as in all-comer populations. Recognizing that drugmakers may not want to risk altering the established indications of their blockbuster immunotherapies with a new biomarker, some companies are betting that sponsors may be more willing to test out new patient stratification models for immunotherapies that haven't yet entered the market.
Nucleai is an example of a company that has made a strategic decision to not advance its deep-learning model for predicting which NSCLC patients are likely to have a durable benefit with Keytruda despite promising results. The firm instead is looking to partner with pharma companies interested in using its spatial biology approach to stratify patients during the early development of next-generation immunotherapies.
Similarly, researchers led by NYU Langone Health's Teresa Davoli and University of California, San Diego's Scott Lippman recently published a study showing that head and neck cancer patients are "exquisitely sensitive" to anti-PD-1 checkpoint inhibitors when they have gene amplifications at the 9p24.1 chromosomal locus. While this biomarker can be gauged by existing NGS tests, the researchers doubted that it would be possible to conduct the studies needed to translate this research into an FDA-approved companion diagnostic without the support of immunotherapy sponsors.
Recognizing the potential of PD-L1/PD-L2 structural variants as predictive biomarkers, Roychowdhury is discussing the possibility of a prospective trial with drug companies that have anti-PD-1 drugs on the market as well as companies developing new treatments. "Based on the prevalence across cancer types, anticipated high response rates, and expected durability, we think this may be an opportunity for tumor-agnostic biomarkers," he said.
In the meantime, his team wants to keep studying the genomic landscape, biological impact, and clinical characteristics associated with these variants. To that end, Roychowdhury has funding from the Gateway for Cancer Research for a telemedicine-based tumor registry to collect samples from cancer patients, genomically characterize the samples, and follow their experience on treatment. A segment of the patients in the registry will likely get on an anti-PD-1 treatment, which will allow researchers to gain additional insights on whether individuals with PD-L1/2 structural variants respond to these treatments.
Since the registry will allow patients to join via telemedicine, researchers will be able to collect data from people living all over the US without making them travel to study sites. Moreover, patients who enroll in the study and provide a sample for analysis will receive $1,000.
In a separate project, Roychowdhury's group is also following up on the finding that PD-L1/2 structural variants occur in the intronic and untranslated regions in these two genes. "Gene fusions are notoriously hard to detect," he said, adding that his lab has designed probes to better gauge these biomarkers via sequencing tumor samples archived at OSUCCC. Roychowdhury is optimistic that he'll not only find more cases with these gene fusions but "find them in the parts of the intron we're not sequencing."