Vokes Voices How to Use Genomics to Predict Response to Checkpoint Inhibitors in Cancer

September 26, 2020

Natalie I. U. Vokes, MD, discusses the rise of predictive genomic biomarkers in oncology.

Although the role of immunotherapy has been revolutionary in the field of oncology, not all patients will respond to treatment, according to Natalie I. U. Vokes, MD, who added that modeling approaches must be used to integrate biology into more accurate predictive biomarkers of response to this approach.

“We're still in the early phases of this research; however, we’re starting to gain insight on the biology of which patients respond to immune checkpoint inhibitors. Better understanding the biology will help with the development of new biomarkers,” explained Vokes. “Even so, we must be cautious in terms of how we apply the biomarkers that we currently have. We need to continue to stress the importance of this biology–biomarker interplay because that will help us treat as many patients as possible with these therapies.”

In an interview with OncLive, Vokes, a medical oncology fellow at Dana-Farber Cancer Institute, discussed the rise of predictive genomic biomarkers in oncology.

OncLive: What is the role of genomics in predicting response to immune checkpoint inhibitors in oncology?

Vokes: The biomarker that is brought up most often is tumor mutational burden (TMB); as a biomarker, I believe this is somewhat limited. TMB is not a strong discriminator between responders and non-responders to this therapy. In clinical practice, when we think about using TMB [as a biomarker], we have to be cognizant of the fact that we're choosing between specificity and sensitivity; there's always going to be that trade-off. However, this doesn’t mean we shouldn't use TMB; we just need to be cognizant when doing so.

More broadly speaking, I believe TMB is interesting. Even though it imperfectly segregates between responders and non-responders, it does track this. I believe this points to the fact that we don't fully understand the biology behind TMB; there are many nuances to that association.

It’s important to remember that there’s a difference between biomarker and biology. We've spent a lot of time trying to jump immediately to the use of biomarkers but there's still a lot that we don't understand about biology. Once we better understand biology, we can better develop these biomarkers. Rather than going straight toward the use of single biomarkers, such as TMB or PD-L1, we could potentially use modeling approaches to integrate this biology into more accurate predictive models [for response].

Could you expand on what is known about TMB? Are any other markers on the rise?

TMB is probably the most well developed in this field. Reflecting back on early studies, many used whole exome sequencing and looked for genomic correlations. These studies indicated that the TMB in patients who responded to therapy was often higher than in those who did not. This was initially demonstrated in the melanoma space. Since then, it has been confirmed with several other sequencing techniques in an array of disease contexts.

Notably, on June 16, 2020, the FDA approved pembrolizumab (Keytruda) for the treatment of patients with unresectable or metastatic solid tumors that are TMB-high in the second-line setting.So TMB is [involved in an important] FDA approval.

Other biomarkers are not as well developed. In the lung cancer space, STK11 mutations come up quite often. However, I believe STK11 is generally a poor prognostic marker across different treatment settings. There's also some other work showing that the presence or absence of other mutations, such as TP53 or KRAS, informs how detrimental the STK11 mutations are, but we still have a lot to learn.

In lung cancer, we tend to use driver mutations as biomarkers like EGFR, ALK, and ROS1; those patients tend to not respond as well [to this treatment]. We also have to be careful that we don't extend this to all driver-altered non–small cell lung cancer because BRAF and MET might be biologically different. There’s still a lot for us to learn about these biomarkers.

What work is being done to effectively predict response to immunotherapy?

We wanted to apply the framework of precision medicine to immunotherapy and, traditionally, that consisted of finding a single genomic alteration or a single gene, which then would allow us to effectively predict whether a patient will respond to therapy. However, those therapies also have a different mechanism.

Immunotherapies are much more complicated, and the landscape of what determines whether a patient is going to respond is complicated. It may not only be determined by the genomics of the tumor itself; we may need to incorporate additional immune features.

With this said, the analysis of single genes is very interesting because it helps us understand how tumor pathways can affect the immune response to the tumor. However, from a biomarker standpoint, we're probably going to need to have some kind of approach that integrates some of these features into a more complicated model.

Where should future research efforts be focused?

Comprehensively aggregating molecularly characterized patient cohorts is a current limitation in this space. I believe working together as a community, bringing together as many samples as we can, and continuing to conduct molecular profiling of these samples is important, as we move forward with this research.

Traditionally, a lot of our work was focused on whole exome sequencing and genomics; however, I believe integrating additional molecular assessment profiles, such as multiplex immunohistochemistry, transcriptomics, and newer approaches, will help enrich the search [for biomarkers] in this space, allowing us to look for more of these associations.

What does the future look like for genomic biomarkers in oncology?

We hope to continue incorporating as many different data types as possible, ultimately, applying more sophisticated modeling techniques. This will help us better understand the biology and will help us build new predictive models. I foresee this being an iterative approach that we can then apply to different data sets in different treatment contexts. Ultimately, this will help us identify some of the differences that exist in terms of how patients respond in the first- and second-line settings.