Immune checkpoint inhibitors have changed the landscape of cancer treatment as a whole. In October 2015 alone, the US FDA approved two therapeutics (Keytruda & Novolumab) targeting PD-1 in metastatic NSCLC, and this effective new treatment mode is starting to provide invaluable options for clinicians and their patients.
At ASCO 2016 I was fascinated to see some of the long-term results of immunotherapy clinical trials which had begun for multiple malignancies and presented at the 2015 meeting. For example, all the pre-conference excitement had focused on Merck & Co’s long term follow-up study to their Phase Ib KEYNOTE trial – the treatment of advanced melanoma patients with Keytruda (pembrolizumab, Anti-PD-1) - which had demonstrated a 33% response, 40% of those responders being alive after 3 years from the start of treatment and a total of 15% experiencing complete remission.
Even though these studies demonstrate real efficacy in reversing dismal outcomes for patients in advanced disease, it has also left behind many questions of biomarker models – which ones, how can we measure therapeutic progress, how do we know what patients are responding and if they are not responding when do we make the decision to change treatment courses? It is clear that the tumour microenvironment is a turbulent, heterogeneous mixture and can differ from patient to patient, making it incredibly difficult to use a single biomarker model. The FDA approval of pembrolizumab and nivolumab with different diagnostic guidelines for use (Pembrolizumab requires the use of PD-L1 IHC 22C3 pharmDx test, the first FDA-approved test to detect PD-L1 expression in non-small cell lung tumours. Whilst Nivolumab is free of such a constraint and the decision to prescribe remains with the physician) despite the similarity of the intended population subset only adds to the confusion.
Notwithstanding the FDA’s acceptance of the PD-L1 diagnostics for Pembrolizumab, the biomarker model continues to proliferate into an ever-expanding matrix of issues for drug developers. Different antibodies, different staining protocols, different target cell assessment (tumour cells, tumour-infiltrating immune cells or both), different scoring methods and thresholds have contributed to a complicated PD-L1 landscape.
“The clinical reality is that some patients respond very well to cancer immunotherapies and others do not. As a result, the ability to leverage our molecular information platform to identify the right candidates for these immunotherapies is an important advance for the field of precision medicine. Matching the right therapy with the right patient has the potential to both improve outcomes and increase efficiency in the current care model.” – Vincent Miller, MD, Chief Medical Officer, Foundation Medicine.
Several talks from the 2016 edition of ASCO tried to address the issue of predicting response using biomarkers:
Luis A Diaz, MD, of John Hopkins University expanded on a phase II trial published last year looking at the treatment of MMR-deficient colorectal cancer by pembrolizumab. He spoke on the abundance of mutant-associated neoantigens resulting in a tumour appearing foreign to the host’s immune system. “[This] foreignness results in an inflamed microenvironment with a very high expression of immune checkpoints. Treatment of these tumors with anti–PD-1 unlocks a potential antitumor response in microsatellite instability tumors,” Dr. Diaz said. Future directions would be to explore the possibility of histology-independent indications for treatment of minimal residual disease tumours with PD-1 blockade, and to investigate the molecular etiology of primary and secondary resistance in tumours with minimal residual disease that have been treated with PD-1 blockade.
Jonathan E. Rosenberg, MD, of Memorial Sloan Kettering Cancer Center, presented results of a trial that examined immune and genetic predictors of response to atezolizumab in metastatic urothelial carcinoma. Predictors of response were identified in PD-L1 immune cell status, Cancer Genome Atlas (TCGA) subtype and mutational load. “Simultaneous assessment of these characteristics may define drivers of immune responsiveness to inform potential combination of strategies,” Dr. Rosenberg said. “These data highlight the importance of the interaction between the tumor and its microenvironment in understanding response to atezolizumab…..The IMvigor201 exploratory biomarker studies suggest that atezolizumab efficacy is driven by genomic, molecular, and immunologic factors related to adaptive immunity.”
Some of the subsequent discussion points for this talk were highlighted by Alexandra Snyder Charen, MD, of Memorial Sloan Kettering Cancer Center. She had noted that those patients not in the highest mutational burden group do not improve as well, and that this subgroup should be the focus on future studies. Furthermore she pointed to some unanswered questions, including why elevated mutation burden matters. The current hypothesis is that mutations generate relevant neoantigens, abnormal proteins, which are then processed and thought to be presented by the tumor or the antigen-presenting cell to the immune system.
Data from other labs suggests that there may be a “few critical neoantigens” scenario, plus “the relative contributions of MHC Class II and tumor-associated antigens in this scenario have not been fully elucidated” she said. It seems for us to get a better understanding of the factors that lead to primary and acquired resistance further studies are needed. “How do we consider tumor heterogeneity and integrate data from the biologic complexity at the tumor site? How do we boil down this complexity into usable biomarkers?” Dr. Charen said.
I have already written about how anti–PD-1 therapy produces a response in about 30% to 45% of patients with advanced melanoma. Douglas Buckner Johnson, MD, MSCI, of Vanderbilt University Medical Center, presented results of a study that used hybrid capture–based next-generation sequencing (HC NGS) to identify markers of response. The sequencing on 236 to 315 genes was performed on archival samples from patients who had been treated with aPD1 to correlate mutational load and specific mutations with clinical outcomes and to compare mutational load as calculated by whole-exome sequencing. The correlation of T-cell receptor sequencing was also examined. In study cohorts, patients will higher mutation load had a response to aPD-1 (median, 45.6 vs 3.9 mutations/megabases; p=0.003). Melanomas with NF1 mutations had a high mutational load, whereas BRAF-mutant, MRAS-mutant, and BRAF/NRAS/NF1 wild-type melanomas had lower median mutational load.
Discussant Padmanee Sharma, MD, PhD, of The University of Texas MD Anderson Cancer Center, pointed out that expression of a single biomarker for selecting patients for treatment “may not be feasible,” given that immune responses are dynamic and evolve over time. He elaborated that studies up to this point “have identified potential prognostic biomarkers, including tumor-infiltrating T cells, immune gene signatures, PD-L1 expression, TcR repertoire, and mutational load, which indicate patients that are likely to have a better response to therapy. A single predictive biomarker has not been identified that can be used to select and/or exclude patients for immune checkpoint therapy, so we have to be very careful how we look at these individually”. Dr Sharma said that “Biomarker development for immune checkpoint agents will require integration of multiple biologic components, such as CD8-infiltrating T cells, PD-L1 expression, IFNγ-signature genes, mutational load, etc., as opposed to a single molecule”.
Dr Yu Sunakawa of the Showa University Northern Yokohama Hospital and Dr. Heinz-Josef Lenz of the University of Southern California, in collaboration with HTG Molecular Diagnostics and Cancer Genetics (CGI) spoke on their study “Immune-related genes to predict clinical outcome of cetuximab (cet) treatment for metastatic colorectal cancer (mCRC): Immuno-Oncology assay research”. ). The success of antitumor activity of cetuximab is based on antibody-dependent cell mediated cytotoxicity mediated by natural killer cells. In terms of progression-free survival (PFS), the study showed that participants could be divided into two groups based on RIPK1 and CEACAM5 genes. For overall survival (OS), patients could be divided into two groups based on the clustering using 7 genes, with OS strongly associated with the IRF3 and BMP4 genes.
Vincent Miller, M.D., chief medical officer at Foundation Medicine, spoke about the use of its genomic profiling technology – FoundationOne – to successfully predict a greater likelihood of response and duration to immunotherapies for patients with advanced bladder cancer and metastatic melanomas. The data suggested that “tumor mutational burden could serve as an independent predictive biomarker to aid clinicians in identifying patients who are most likely to benefit from cancer immunotherapies that target either the PD-1 or PD-L1 proteins” he said. With reference to this solution Vamsidhar Velcheti, M.D., assistant professor, Solid Tumor Oncology, Taussig Cancer Institute, Cleveland Clinic, said that “these results are particularly exciting given the amount of variability inherent to using immunohistochemistry (IHC) to measure biomarkers. There are many different PD-L1 IHC tests, for example, and pathologists often do not see agreement between them. We need a truly quantitative and reproducible approach to predicting response to immunotherapies, and measuring tumor mutational burden using FoundationOne may provide us with that solution.”
According to a study involving Laura Q.M. Chow, MD, associate professor of medical oncology at University of Washington School of Medicine and associate member of clinical research at Fred Hutchinson Cancer Research Center, the use of PD-L1 expression as a biomarker on tumor cells and inflammatory cells may predict which patients with recurrent or metastatic head and neck squamous cell carcinoma will benefit from treatment with pembrolizumab. Furthermore, biomarkers other than PD-L1 may also better define the tumor microenvironment and predict response to pembrolizumab (Keytruda, Merck).
Laura Chow said “This is the first full study looking at biomarkers that incorporate inflammatory cells with PD-L1 expression, PD-L2, as well as interferon-gamma gene signatures. By looking at all of these different biomarkers, we were able to use a multivariate approach to finally better define the tumor microenvironment and better predict who is going to respond to pembrolizumab.” Chow and colleagues conducted analysis of the potential markers – PD-L1, PD-L2 and interferon-gamma – as they correlated with the patient response in phase 1 of the KEYNOTE-012 study.
Researchers used immunohistochemistry in tumor cells alone (tumor proportions score) or with inflammatory cells (combined proportions score) to determine PD-L1 and PD-L2 negativity and positivity in baseline tissue samples. On the results, Chow stated that “It appears that PD-L1 seems correlate with PD-L2, and there is a significant association between PD-L1 and PD-L2 expression”.“There is some difference between our study and the CheckMate 141 study, and these data need to be further explored, especially in terms of whether we should be incorporating inflammatory cells into the PDL-1 assay and what antibodies and cutoffs should be used. A lot of these answers are still not entirely available.”
With all of the excitement and data to come out of ASCO 2016 surrounding biomarkers, it is clear that the field is steering towards the need to include biomarkers into their future clinical strategy. There has been huge investment in this part of the drug pipeline and the degree of its success is linked to accurately predicting immune responses in different patients; especially given the small patient subsets that have benefited in clinical trials to data (33% of patients showed a response in Merck phase 1b KEYNOTE Trial). The key question is what studies and investment needs to be put in to understand the complexity of the tumor microenvironment and pinpoint clinically-applicable biomarkers? It has become apparent that a single predictive biomarker will not be feasible given that immune environment is so dynamic and further investigations will be made into arrays of biomarkers. As precision medicine enters into the limelight, and infrastructure builds for proteogenomic databases, it will be interesting to see what biomarkers are recognized as clinical standards. Although the industry has advanced a significant amount there are still a lot of questions to be answered and further avenues to be explored. This is definitely an area to keep our eye on and I look forward to hearing the results from planned studies in ASCO 2017.
Guest blogger: Peter Harkness is passionate about the future of medicine and has been active in the life science and healthcare sectors for 10 years. He has a particular interest in Cell, Gene & Immunotherapy and areas where medicine and technology intersect. He has worked at numerous media organisations including The Economist. Currently he is a guest blogger for Phacilitate.