Contributed Commentary by Natalie LaFranzo
February 7, 2020 | Achieving true precision medicine in oncology continues to challenge the entire healthcare field—from clinicians to drug manufacturers to diagnostics companies. Cancer has been described as a “continuous disease process”, which cannot be defined using one single morphology or mutation. Recognizing this, the field has incorporated comprehensive molecular profiling of tumors into the standard of care, and targeted therapies have been developed based on this information. However, despite the billions of DNA sequences that have been generated, this data is sifted through to identify individual mutations that are used for downstream diagnostics to define a patient’s treatment path. This approach fails in two regards:
- DNA is static. It captures the risk of developing a disease, but does not tell us whether environmental or other factors have “turned on” these signals, thereby causing disease to manifest or progress.
- Single mutations or other independent analytes cannot capture the complexity of disease. Yet, we continue to try to characterize patients based on these discrete, individual analytes rather than in the biological concert in which they exist.
The impact of these failures is significant. From a drug development and approval perspective, not understanding the target patient population can result in delayed trials, or even worse, trial failures and drug abandonment. From a patient perspective, mismatched treatment plans are costly, dangerous, and disappointing for patients.
The most recent advancement in this area has come with the introduction of tumor mutational burden (TMB) as a putative biomarker for immunotherapy. While this approach addresses the individual analyte challenges described above, it still relies on the static nature of DNA. And, as a result, despite early promising results, TMB has not delivered on the hype.
In order to achieve this goal of truly tailoring therapy regimens to patient’s molecular profiles, it’s imperative that we overcome both of the challenges above and move into the next era of biomarker and diagnostic development.
Why Predictive Immune Modeling?
The field of Predictive Immune Modeling addresses these legacy challenges and serves as the infrastructure on which the future of diagnostics will be built, for oncology and beyond. Predictive Immune Modeling uses RNA data as the foundation for characterizing disease and patient populations. RNA expression, unlike DNA, is dynamically modulated in response to stimuli such as diet, exercise, disease state, and therapy. This enables a more real-time assessment, in particular, at the site of the tumor. Historically, RNA has been considered “difficult” to work with, due to its unstable nature, as well as the degradation and chemical modifications that occur when preserving samples for later analysis. But, advances in molecular and analysis tools for high-throughput sequencing have allowed us to better characterize RNA from these difficult samples, resulting in a massive boom of RNA data.
However, big data has utility only when it’s distilled into meaningful signals, suitable for clinical decision-making. Unlike in the DNA space, where large data sets have been simplified down to single-analyte signals, Predictive Immune Modeling uses RNA models to capture complexity in a meaningful way. These RNA models represent not only the presence or absence of RNA, but also the dynamic expression level. To date, these models have been built to characterize clinically-meaningful biological systems, such as immune Health Expression Models (iHEMs). These models enable signals such as immune composition or cell state to be measured directly from RNA extracted from a solid tumor—making this approach amenable to samples with minimal tissue, common for patient materials in the clinic.
Certainly, these cell signals do not operate in isolation or independent of one another, as they represent the dynamic immune response happening at the site of the tumor. Predictive Immune Modeling uses machine-learning to look at the concert of immune signals in the tumor microenvironment. After a clinical trial has been completed, patients are grouped by response criteria such as Response Evaluation Criteria in Solid Tumors (RECIST). The immune profiles of these cohorts serve as input to create a multidimensional biomarker that will predict future patients’ response to therapy. Diagnostics built to differentiate responders and non-responders [to a therapy] using this approach capture significantly more information about the dynamic immune response—enabling better predictive ability.
The Future of Diagnostics
Looking forward, there is excitement around how the discipline of Predictive Immune Modeling overcomes many of the historic challenges which have limited the impact of oncology diagnostics to date. The field is in agreement that moving away from single-analyte, static biomarkers and into the new world of dynamic, multidimensional biomarkers will enable better drug development and for improved patient treatment decisions. Importantly, as new clinical targets are discovered, Predictive Immune Modeling allows for the development of new RNA models, or integration of additional data. The future of precision medicine is founded in comprehensive characterization and signal prioritization through machine-learning, and this is nowhere more evident than in the evolution of immune profiling for predictive diagnostics.
Natalie LaFranzo, PhD. supports Cofactor Genomics’ clinical collaborations and outreach efforts, with the goal of validating and expanding the utility of multidimensional RNA biomarkers. Previously, she developed customized experimental solutions for both DNA and RNA applications at Cofactor, as well as launched and supported diagnostic reference standards as a part of Horizon Discovery’s Diagnostics Division. She can be reached at natalie_lafranzo@cofactorgenomics.com.