Immune checkpoint inhibitors (ICIs) have revolutionized the cancer therapy landscape from evidence-based medicine to precision medicine (BMC Cancer 2021; doi: 10.1186/s12885-021-08662-2). Over the past decade, ICIs targeting the programmed cell death protein-1 (PD-1)/programmed death ligand-1 (PD-L1) axis have emerged as standard-of-care treatments for non-small cell lung cancer (NSCLC) (Curr Oncol 2018; https://doi.org/10.3747/co.25.3750).
Although ICI therapy achieves long-term survival in some patients, only a small fraction of patients actually respond to this treatment strategy, and many of those who derive initial clinical benefit experience disease progression at a later stage. The overall response rate is modest, ranging from 20 percent to 50 percent in patients with NSCLC (Curr Oncol 2018; https://doi.org/10.3747/co.25.3750).
To maximize the likelihood of selecting the correct therapy combinations for each patient, clinicians rely on predictive biomarkers to guide individual treatment planning. However, a key reason for precision medicine falling short for cancer patients is that current biomarkers for immunotherapy outcomes have low predictive power, and the mechanisms behind treatment resistance are not fully understood (Clin Cancer Res 2019; https://doi.org/10.1158/1078-0432.CCR-18-1538).
To tackle this issue, a recent study published in The Journal for ImmunoTherapy of Cancer, highlights the complex tumor-patient-treatment biologic interplay and the role of blood plasma proteomic profiling for assessing resistance in NSCLC patients being treated with ICIs (2022; doi: 10.1136/jitc-2022-004582).
In the study, researchers aimed to identify blood-based proteomic biomarkers for response to ICI therapy by investigating pre-treatment and on-treatment plasma proteomic profiles of patients with advanced stage NSCLC. Overall, 108 subjects participated in the study, of whom 80 were responders and 28 were non-responders.
Utilizing a first-of-its-kind advanced machine learning and bioinformatics approach, roughly 800 pre-treatment and on-treatment plasma proteins from 143 ICI-treated patients with NSCLC were screened. Clinical parameters collected from patients included specific mutations, body mass index, and smoking habits.
The researchers discovered a predictive signature for response that consists of two proteins-CXCL8 (IP-10) and CXCL10 (IL-8)-and two clinical parameters (age and sex). The signature yielded a receiver operating characteristics (ROC) area under the curve of approximately 0.8. In addition, exploration of the proteomic profiles revealed three patient clusters with distinct clinical and biological features. Each subtype correlated with distinct multiple clinical and functional characteristics such as sex, response, and TNM (tumors, nodes, and metastasis) staging.
An enrichment test to explore the trends in proteomic patterns revealed that neutrophils play a role in ICI therapy resistance. Patients who did not benefit from ICI therapy were associated with on-treatment changes in neutrophil function or enrichment in neutrophil subsets. The authors report that this is the largest and most comprehensive plasma proteomic data set for ICI-treated patients with NSCLC available to date , at baseline and on treatment, as previous plasma biomarker studies investigated a limited number of proteins.
Oncology Times chatted with Yuval Shaked, PhD, Professor in the Rappaport Faculty of Medicine, Head of the Technion Integrated Cancer Center, President of the International Cancer Microenvironment Society, and lead author of the study. He provided in-depth analysis of his team's unique approach to analyzing patient-specific resistance mechanisms using proteomic analysis and enabling personalized treatment options for patients with NSCLC.
Oncology Times: ICI-based treatment has revolutionized the cancer therapy landscape. However, only a small fraction of patients actually respond to this type of treatment strategy. What are the current challenges when it comes to predicting treatment response in NSCLC? Should there be a change in what the perception of precision medicine should be?
Shaked: "Identifying patients who will benefit from immunotherapy is still a major challenge in clinical oncology. It turns out that most patients do not respond to treatment, and some even display hyperprogressive disease while on treatment. The biomarkers used in the clinic today for guiding treatment decisions are based on the expression level of PD-L1 in tumors, as well as tumor mutational burden. These biomarkers have very modest predictive power. In addition, the biomarker tests require tumor tissue which is obtained via invasive procedures. But the challenges do not end here. Lack of good predictive biomarkers means inefficient resource allocation, unnecessary adverse events and, most importantly, waste of time for patients who have no time to spare.
"In many cases, following non-response to first line, the next line choices are based on trial and error. Better biomarkers are desperately needed. In fact, I believe that there needs to be a fundamental change in how we perceive precision medicine. We should strive to develop precision medicine solutions as a tool for patients, clinicians, and payors, and not as a tool for biopharma. We know that no two patients are identical and, as we develop precision medicine tools, one should try to improve the resolution of clinical assessment.
"Think about imaging as an example, from the X-ray of the early 20th century to the CT of the '70s and later, the MRI, and modern imaging tools. Better resolution has better clinical insight. In our study, we demonstrate such an approach. Combining proteins' pattern analysis in the plasma with machine learning and bioinformatic tools, we were able to identify a 'signature' that not only outperforms current biomarkers, but also provides interesting insights on the potential resistance mechanisms that are active and hindering response to treatment.
"For example, we found an enrichment of neutrophil-related proteins in patients resistant to therapy. Our findings not only provide a basis for identifying patients who are likely to respond to immunotherapy, but also shed light on the potential mechanisms that promote response or resistance to therapy."
Oncology Times: What potential value does this machine-learning and bioinformatics-based approach provide to both patients with cancer and clinicians?
Shaked: "Our approach provides several tiers of information that support clinical decision-making. First, it provides a tool for predicting whether the patient is likely to benefit from the treatment or not. Second, it identifies biological processes and proteins potentially driving treatment resistance or response. Third, identifying potential mechanisms of resistance may then guide the search for drugs that may be combined with immunotherapy to overcome therapy resistance. For example, if angiogenesis is identified as the potential mechanism of resistance, adding an anti-angiogenic drug to the backbone immunotherapy could be a potential treatment strategy for the patient."
Oncology Times: What follow-up studies need to be performed to clarify the true clinical value of this prediction model? Are there any plans to improve the current technology capabilities of this approach?
Shaked: "As our study was conducted on a retrospective cohort, we are now aiming to demonstrate the power of our approach in prospective studies. We are currently working on large, multi-center, prospective studies in which we measure up to 7,000 proteins in the blood in different cancers, including NSCLC and melanoma. This prospective, multi-center clinical trial is recruiting patients at over 40 clinical sites in the U.S., U.K., Europe, and Israel. Over 700 patients have been recruited so far, and we have been able to validate and improve the capability of our platform to predict response, analyze resistance mechanisms, and associate those with either approved medications or ongoing clinical trials.
"We plan on launching our first commercial product later this year in the U.S. and publishing the results of the trial in a peer-reviewed journal later this year. Interim results were announced in January."
Oncology Times: Are these findings widely applicable and is there a potential of transferability of this combination of artificial intelligence machine learning with proteomic profiling to predict treatment response in other cancers? Are there additional clinical trial sites/partnerships opening around the world to expand the research?
Shaked: "Indeed, the approach can be applied to additional cancers and different treatment strategies. We are currently focusing on immunotherapy for the treatment of NSCLC and melanoma via ongoing, multi-center, prospective studies, and plan to expand to more indications and treatments in the future. It is most probable that unique signatures will be identified for each indication/treatment, although some features may be common across indications. Overall, we believe that our study sets the stage for identifying powerful predictive biomarkers based on liquid biopsies, advancing precision medicine for cancer."
Dibash Kumar Das is a contributing writer.