Skip to main content

Multi-Biomarker Strategies Show Promise in Improving Immunotherapy Response Prediction


NEW YORK – Researchers have identified new strategies for predicting response to immunotherapy, using approaches that integrate genomic and molecular signals, immune cell phenotypes, and other biological markers related to the body's anti-tumor immune response.

Although further validation is needed, early data on two multi-parameter strategies presented this week at the European Society for Medical Oncology's Virtual Congress — one exploring blood-borne signals and another using DNA and RNA-based signatures — suggest they may improve oncologists' ability to identify which of their patients will benefit from immune checkpoint inhibitors (ICI).

The concept that some sort of combinatorial strategy might be necessary for more precise prediction of immunotherapy response is not new. Almost five years ago, when ICIs had just entered the market, a group from the Netherlands Cancer Institute and the University of California, Los Angeles proposed a multi-pronged approach they called a "cancer immunogram" for assessing a patient's likelihood of responding to treatment.

But in practice, immunotherapies have largely advanced either without any companion diagnostic biomarkers, or with specific, single-parameter tests that gauge PD-L1 expression, microsatellite instability, or tumor mutational burden (TMB) — all of which are recognized as capturing only some of the variability in immunotherapy response, and as such, potentially missing responders or mistakenly identifying non-responders.

"Immune checkpoint inhibition can induce deep and durable responses in many cancer types. However, we are still struggling to identify patients that may or may not benefit," Daniela Thommen of the Netherlands Cancer Institute's division of molecular oncology and immunology said during a discussion of the two studies presented at ESMO.

The biomarkers that doctors are currently using in the clinic are not without value, she noted. PD-L1, for example, has proven useful for directing treatment when patients are exceptionally high expressers. And TMB has been shown to associate with response across many tumor types, resulting in the recent histology-agnostic US Food and Drug Administration approval of pembrolizumab (Merck's Keytruda) as a treatment for adult and pediatric patients with refractory solid tumors and a high mutational burden.

However, Thommen said, "both markers are still far from perfect. For instance, patients with high PD-L1 and high TMB may not respond … and vice versa, patients with low PD-L1 expression and low TMB can still benefit."

The limitations of PD-L1 and TMB have spurred the search for better immunotherapy biomarkers, some of which were highlighted at ESMO. One of these strategies was discussed by Kevin Litchfield at the University College London Cancer Institute, whose team conducted a meta-analysis using publicly available exome and transcriptome data from more than 1,000 patients with various tumor types enrolled in multiple immunotherapy studies.

Investigators reanalyzed this data using a standardized bioinformatics pipeline to identify various signatures that captured cancer cell states, tumor immune inflammation, and immune inhibitory pathways. According to the authors, two parameters — clonal TMB and the gene CXCL9 — were the strongest predictors across the dataset, beating out other putatively predictive factors including various established gene-expression-based scores of tumor immune response and immune cell infiltration.

The fact that TMB and CXCL9 performed the strongest in this analysis interested Thommen since both are crucial for a tumor-specific T-cell response. Clonal TMB contributes to clonal neoantigen presentation and CXCL9 is a "critical mediator of T-cell infiltration," she noted.

Also notable in this research was the fact that certain markers that have previously been reported to have strong predictive power completely failed to replicate in this meta-analysis, including the burden of somatic copy number alterations.

Litchfield and colleagues noted there was significant correlation between certain biomarkers, indicating that there probably is redundancy between all the different predictive factors identified. But it was clear that certain factors have minimal correlation with one another, and thus, might have added value when combined.

Based on this hypothesis, the researchers fed the most predictive parameters through a machine learning pipeline, deriving a multivariate score that showed consistently high predictive value, even when applied in an independent cohort. They reported that this score outperformed TMB across a number of tumor-specific datasets, including pan-cancer samples, lung cancer, and melanoma.

Thommen said that the standardized data processing and outcome harmonization that Litchfield and colleagues used lend credence to the results. She also highlighted the fact that the UCL team's score was able to identify patients benefitting from ICI treatment even if they had both low TMB and negative PD-L1 results.

But she said it would be valuable to see the UCL team's combination signature tested head-to-head against other published methods that combine TMB with T-cell or immune response gene expression signatures.

In a second ESMO presentation, Giulia Mazzaschi, a researcher in the University of Parma's medical oncology unit, shared results of her team's study exploring various circulating factors that are implicated in the immune response to cancer. The team explored whether the combination of these individual markers detected in blood samples of advanced non-small cell lung cancer patients may predict their responses to ICI treatment.

The team collected peripheral blood at baseline from 109 NSCLC patients undergoing ICI treatment between August 2015 and April 2019. At the time of data cutoff and after a median follow up of 17.8 months, median progression-free survival and overall survival was 2.6 and 7.9 months, respectively.

For each patient, the researchers quantified the presence of different circulating cells, calculated what is known as a derived neutrophil to lymphocyte ratio (DNLr). They also analyzed biochemical parameters, including LDH levels, which allowed them to measure an established prognostic algorithm called the Lung Immune Prognostic Index (LIPI), and tested soluble PD-L1 using immunoassays.

Finally, they counted CD8-positive and PD1-positive cells, and NK cells using flow cytometry, which they integrated into a novel immune effector score, dubbed LeffS.

Looking at each factor individually, the group saw that all showed a correlation with ICI treatment response. A DNLr greater than 3 was clearly associated with poor outcomes, as were higher LDH levels.

Looking at soluble PD-L1, low levels conferred a significantly better progression-free survival and overall survival compared with high levels, something the Parma team has seen in previous research. There were also significant increases in both CD8/PD1-positive cells and NK cells among the subset of patients with a clinical benefit from ICI.

Combining both cell quantitation and soluble PD-L1 into the LeffS score improved risk prediction, and provided a stronger association than the LIPI, with an AUC of 0.80. Patients with a favorable LeffS has significantly prolonged progression-free survival and greater benefit from immunotherapy treatment, Mazzaschi said.

Finally, the team combined both LIPI and LeffS and attempted to divide the cohort into three response groups with distinct outcomes. Individuals with zero or one risk factor had significantly prolonged progression-free and overall survival on ICI compared to those with two or three factors, while those with four or more factors fared the worst.

"The impact of combined integration was striking also in terms of tumor response," Mazzaschi added. "None of patients with the poorest scores responded to ICI, while 60 percent of the clinical benefit group displayed zero or one risk factor."

Discrimination of response and non-response using the full multi-parameter combination was clearly better than either method alone, with an overall AUC of 0.90.

The team is now working to validate these results in an independent cohort of NSCLC patients, and to explore even broader approaches incorporating radiographic measures and other immune and multi-omics signatures.

According to Thommen, the fact that the group's method is blood based is a big plus for potential clinical implementation. But in addition to further validation, she said the Parma team will also need to make sure they understand how their score might be influenced by either systemic inflammation or infectious diseases, as both could likely alter many of the included parameters.