In a paper published in JAMA Oncology, researchers from Johns Hopkins University conducted a systematic review to compare the efficacy of different biomarker modalities in predicting response to PD-1/PD-L1 checkpoint blockade.
The researchers found that multiplex immunohistochemistry/immunofluorescence (mIHC/IF) performed significantly higher in measures of diagnostic accuracy for predicting anti-PD-1/PD-L1 therapy when compared to other techniques, such as PD-L1 IHC, tumor mutational burden (TMB), and gene expression profiling.
We explored these findings in-depth in a webinar featuring the first author of the paper, Steve Lu, an MD/PhD candidate from Johns Hopkins University and Dr. Cliff Hoyt, VP of Translational and Scientific Affairs at Akoya Biosciences.
Cancer immunotherapy and the tumor microenvironment
In recent years, we’ve seen lasting benefits resulting from cancer immunotherapies, but many patients don’t respond to treatment. Effectively classifying patient populations is important to ensure the right patients receive the right treatment.
“The goal [for] us in the biomarker space is to develop tests that help further stratify patients into subclasses, within which you get higher response rates,” said Hoyt.
The prevailing thinking, he added, is that the answers lie in the tumor microenvironment (TME). Cancer cell behavior is a function of what is happening in their surrounding environment, including the signals they receive from other cells.
The only practical way to observe the TME is to stain tissue sections with multiple markers and image. Multiplex staining offers more information and context about the microenvironment compared to single stain methods, said Hoyt. You can distinguish between cell types, observe their functional states, and identify any checkpoint-related proteins present.
Multiplex IF also supports our understanding of the cell-to-cell biology driving disease progression. With spatial context, it’s possible to unlock cellular mechanisms behind disease and discover new therapeutic targets. This can be used in trials to confirm drug mechanism of action and identify patient cohorts enriched for response – especially helpful for pharmaceutical companies in order to develop effective therapies.
At Akoya, our goal is to provide platforms that support the biomarker development continuum from discovery to clinical commercialization. Hoyt emphasized that requirements for these platforms change as you travel along the continuum. Research in the discovery phase requires platforms like CODEX® , which have high multiplexing capabilities to support biomarker discovery. Simplicity is key at the clinical end – imaging systems like the Vectra® and Mantra® offer lower-plex but higher throughput.
Evaluating different biomarker strategies
In their paper, Lu and his colleagues evaluated four different categories of biomarker strategies: PD-L1 IHC, tumor mutational burden (TMB), gene expression profiling, multiplex IHC/IF (mIHC/IF), and multimodal strategies which combine multiple biomarker approaches.
The key question they sought to answer, Lu said, was what the relative efficacies of these different biomarker approaches were for predicting response to anti-PD-1 monotherapy.
Lu and his team conducted a series of analyses to answer this question. For each individual study they looked at, they extracted the patient outcomes and correlated them with test results. They could then compare a patient’s responder or non-responder status to their assay test results. The sensitivity and specificity for each individual study were plotted.
In order to evaluate how each biomarker strategy was performing relative to the others, Lu’s team generated summary receiver operating characteristic (sROC) curves. From this, they were able to calculate the relative area under the curves (AUCs) as a measurement of each approach’s ability to distinguish responders from non-responders.
The AUC of the mIHC/IF approach was significantly higher than that of other modalities. This held true when the studies were weighted equally and when they were weighted by study size. The performance of the other approaches, PD-L1 IHC, TMB, and GEP, did not significantly differ from one another.
Lu’s team also looked at multimodal approaches which incorporated multiple tests. The AUC of these approaches was also higher than the non-mIHC/IF approaches alone, and not significantly different from mIHC/IF. However, with multimodal approaches, said Lu, it’s important to weight the ease of implementation. In comparison to a simple PD-L1 IHC analysis, a multimodal approach will be more time-consuming and expensive.
The fact that multimodality produced a higher AUC than the non-mIHC/IF approaches alone suggests that the single approaches are measuring different things. An inflamed TME, which is measured by PD-L1 IHC or GEP may be distinct from tumor antigenicity, measured by TMB. For this reason, combining these tests may lead to valuable insights.
The researchers also calculated positive and negative predictive values for each study. While most modalities had high negative predictive values, mIHC/IF also had high positive predictive values. This suggests that mIHC/IF is better at identifying patients who would benefit from treatment.
Likelihood ratios (LRs), a value that combines information about sensitivity and specificity, were also generated for the studies. The LR indicates the probability that a patient will respond to treatment based on their test result. Analysis showed that mIHC/IF has better, more useful negative and positive LRs compared to other modalities.
Across multiple measurements used in their meta-analysis, Lu and his team found that multiplex IHC/IF consistently demonstrated better performance.
“By staining multiple markers on fewer slides, you’re able to get co-expression information, spatial relationships, and also proximity and distance information about the tumor microenvironment,” said Lu. “It’s not unreasonable to expect that higher resolution of the tumor microenvironment would likely lead to more insight about a patient’s response to therapy.”