Where it all began
On September 4, 2014, pembrolizumab became the first PD-1 inhibitor to receive approval for patients with advanced or unresectable melanoma[1]. This was followed by the broadening of the clinical development of PD-1 and PD-L1 inhibitors as anticancer agents and in early 2015, the FDA approved nivolumab for the treatment of squamous cell non-small-cell lung cancer (NSCLC).
In October 2015, the anti–PD-L1 agent atezolizumab earned approval for use in patients with metastatic NSCLC that has advanced in spite of first-line treatment with chemotherapy. For patients with EGFR or ALK mutations, the drug is indicated for use after the disease has progressed with an FDA-approved targeted therapy[2].
Anti-PD-L1 treatments work by inhibiting a protein receptor called PD-1 on T cells, a type of immune cell. PD-1 belongs to a family of checkpoint proteins that, when activated, serve as a brake on the immune system. These treatments prevent tumor cells from communicating through the PD-1 protein to inactivate T cells, allowing the immune system to attack the tumor cells.
But who would respond to treatment? The idea was simple. Patients would be tested for PD-L1 using immunohistochemistry, an assay which stains tissue specimens according to the presence or absence of a biomarker, and how much of that biomarker is expressed.
As years have passed and countless patients have been treated in this way, we have seen that for many tumor types, PD-L1 actually has very low predictive capabilities, but rather prognostic. Patients who have high PD-L1 levels and are expected to benefit from treatment, do not; Patients with low PD-L1 levels have shown to have good clinical outcomes.
The big clinical question
The PD-L1 biomarker has limited predictive value and does not direct physicians on whether they should treat the patient with immunotherapy or not, and in some cases, the guidelines aren’t clear. For instance, if a patient has a PD-L1 >=50%, there is no guidance as to whether to administer monotherapy or combination therapy.
Physicians confront multiple predicaments occurring simultaneously when choosing a treatment plan for their lung cancer patients, finding themselves in a position of great uncertainty as they face a process with very limited support. With the current “one-size-fits-all” protocol, patients begin either targeted therapies or immunotherapy, alone or in combination with chemotherapy, and if both traditional and available biomarker-informed treatments fail, some patients are referred to clinical trials – as long as it isn’t too late.
This is what we have in the market today – a lot of unanswered clinical dilemmas that physicians face on a daily basis. It has fast become acknowledged that we need a biomarker that goes beyond what PD-L1 offers.
The linear approach is not enough
The fact is that the majority of NSCLC patients fail to clinically benefit from checkpoint inhibitors, notably antibodies targeting PD-1 and PD-L1, and uncertainties remain regarding how best to use these therapies in clinical practice. Response rates for metastatic NSCLC treated with immunotherapy still barely reach 30% – meaning that, on average, only about three of every 10 patients will benefit from treatment over time.
Given the risk of immune-related and other adverse effects associated with treatment, there is an urgent need to identify biomarkers that can accurately predict which patients will benefit and which will not. If we can discover biomarkers that tell us earlier what the right treatment might be and what response trajectories might look like, clinicians can make informed decisions, identify resistance, intervene sooner, closely monitor patients with a high risk of resistance, and choose next-line therapies based on patient and cancer biology, rather than relying on one-size-fits-all protocols.
While we would expect the linear association between a drug and a biomarker to work, it clearly not does not.
Why?
Linear association very rarely exists in biology, specifically in clinical samples, and despite our deep desire to pinpoint it, it has proven to be extremely difficult. In fact, linear association is the key issue we face when trying to use old paradigms for developing biomarkers and it very rarely exists in cancer. The linear association approach is in line with the traditional drug-target identification process – it works well when we look for targets, but it is less effective when we look for biomarkers. This is because the approach neglects the fact that the interaction between the therapy and the tumor is taking place within a very sophisticated biological system with many ongoing processes in parallel – the patient.
There are multiple co-factors that affect response, both clinical and biological, and relying on a single biomarker oversimplifies the biological factors. When it comes to biology, it’s very difficult to find a single silver bullet to explain the entire variability in the complex tumor tissue, let alone the interaction with the host – the patient themselves.
There is a continuous interaction between the patient, tumor, and therapy – so, during biomarker discovery, we need to consider a complex dynamic system that differs both between patients and within the same patient and changes over time. This is not ideal when trying to identify and develop robust biomarkers. Access to tissue is also a challenge; tumor tissue is not always available or usable, limiting lab professionals’ ability to look for new biomarkers.
When working to discover new biomarkers to address this issue of complexity and multidimensionality, we need to take a different approach, and the key strategy is to look at the bigger picture, try to add to this failing linear association. The truth is that additional biological features tell us a much more comprehensive story.
While DNA and RNA analyses have been the cornerstone in current biomarker discovery, proteins are the functional entities in the cell, and provide us with a holistic view of what is taking place biologically inside the patient’s body. Proteins give us deep insight into the complex interplay between the patient, the tumor, and the treatment, increasing the odds of identifying a clinically insightful biomarker platform. However, proteins also pose a significant challenge. High variability of protein levels and expression between patients, overlap between different proteins and different biological processes, and dynamic everchanging expression are only some of the challenges of using proteins as a biomarker development platform. To deal with so many features, complex interactions and dynamic modeling technology is needed. Machine learning tools and pattern recognition capabilities combined with biology and bioinformatics are the tools we need to characterize such complex systems.
We have developed a technology platform that combines these tools in order to develop a “hybrid’ biomarker to outperform PD-L1.
Improving on PD-L1 with PROphet® NSCLC
Here at OncoHost, we look at resistance as a whole; PD-L1 is just one piece of the puzzle.
Our aim was to try and understand resistance dynamics and biology in a more comprehensive way. So we developed PROphet®, a first-of-its-kind plasma-based proteomic pattern recognition tool that combines system biology, bioinformatics, and machine learning to provide clinicians with actionable clinical insights, optimal therapy choices, and a better understanding of their patients’ personalized cancer dynamics.
PROphet® is developed following strict data science best practices, with clinical data collected from our ongoing multinational, multicenter clinical trial, PROPHETIC.
Our first test, PROphet® NSCLC, identifies expression patterns in a panel of approximately 7,000 proteins using patient’s blood. So, instead of measuring just one biomarker, we measure 7,000 potential biomarkers. As it is a very complex process to get a clear answer from 7,000 biomarkers, we looked for patterns in different cohorts from our clinical trial and identified various proteomic patterns for specific segments of patients.
The result?
The predictive power of those patterns, the outcome of differentially expressed proteins in individual patients, gave us a tool that is much more accurate than PD-L1 in terms of its ability to predict clinical benefit. We can now provide real clinical utility for informing treatment decisions for NSCLC patients by adding resolution to the PD-L1 biomarker, enabling selection of the most suitable treatment modality for each patient.[3]
Requiring just a single, pre-treatment blood sample, PROphet® delivers a report that predicts a patient’s clinical benefit (Progression-Free Survival>12 months) from anti-PD-1/PD-L1 immunotherapy-based treatment plans. Combining these findings with a patient’s PD-L1 level allows for a clear distinction between patients who will benefit from immunotherapy alone versus immunotherapy combined with chemotherapy. In addition, it may improve the patient’s overall response rate. PROphet® thereby addresses one of the most common daily dilemmas of the oncologist with an accuracy and level of resolution that simply does not exist today.
With PROphet®, physicians can offer the most effective plan for each individual patient, avoiding unnecessary treatments and their potential toxicities and enabling further refinement of current guidelines. We hope to create a shift in the industry and improve the lives of those fighting this disease.
Editor’s Note: Dr. Ofer Sharon is a physician and entrepreneur with over two decades of experience in clinical research, pharmaceuticals, and biotechnology, and has made vital contributions to the acceleration of personalized medicine and oncology drug development. He is the CEO of OncoHost, a precision diagnostics company centered on predictive biomarker development for improved patient care.
[1] https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5778665/#:~:text=Melanoma-,Pembrolizumab,study%20%5B13%2C%2014%5D
[2] https://aacrjournals.org/cancerdiscovery/article/6/12/OF1/5424/First-Anti-PD-L1-Drug-Approved-for-NSCLCFirst-Anti
[3] https://www.medrxiv.org/content/10.1101/2022.12.01.22282769v1
