NEW YORK (GenomeWeb) – A team led by researchers at the University of Maryland and the National Cancer Institute has developed a gene expression-based predictor of response to immune checkpoint blockade therapy in metastatic melanoma patients.
In a study published yesterday in Nature Medicine, the scientists, led by senior author Eytan Ruppin and first author Noam Auslander, both of the Center for Bioinformatics and Computational Biology at the University of Maryland and the Cancer Data Science Lab at NCI, described their so-called immuno-predictive score (IMPRES), which can predict response of melanoma patients to checkpoint inhibitors with better accuracy than existing approaches.
"There is a critical need to be able to predict how cancer patients will respond to this type of immunotherapy," said Ruppin in a statement. "Being able to predict who is highly likely to respond and who isn't will enable us to more accurately and precisely guide patients' treatment."
Initially, the researchers focused on neuroblastoma, which often shows spontaneous regression in young children, mediated by cellular immunity. An immune-based predictor of such spontaneous regression, they reckoned, might also predict response to checkpoint inhibitors in melanoma patients.
Using transcriptomics data from 108 neuroblastoma patients, including some who spontaneously regressed and others who progressed, they built a predictor that focused on the expression of 28 immune checkpoint genes and considered pairwise relations between the expression levels of these genes. The result, IMPRES, was able to predict spontaneous regression in the neuroblastoma dataset with high accuracy.
Next, they looked at nine melanoma gene expression datasets and found pathways that were consistently differentially expressed between checkpoint inhibitor responders and non-responders.
They then calculated IMPRES scores for 256 melanoma patient samples from nine datasets from six independent studies, which included patients treated with anti-CTLA-4, anti-PD1, and a combination of the two, and found that the scores accurately predicted the patients' responses. They also tested IMPRES on a new transcriptome dataset from tumor biopsies of 31 metastatic melanoma patients who were treated with checkpoint inhibitors, almost 300 samples in total, and found that it predicted almost all true responders and misclassified less than half of non-responders.
In addition, they found that IMPRES performed better than other types of predictors, such as cytolytic activity and PDL-1 expression.
They also trained a predictor of checkpoint inhibitor response based on melanoma data instead of neuroblastoma data and found that its performance was lower, suggesting that it is important to use truly independent training data.
Overall, they concluded, the good performance of IMPRES results from the fact that immune mechanisms that are correlated with spontaneous regression in neuroblastoma are also involved in response to checkpoint blockade therapy in melanoma and that those mechanisms can be captured by expression differences between pairs of immune checkpoint genes.
"Future studies are warranted to further study the predictive performance of the approach presented here in other cancer types for which [immune checkpoint blockade] is approved as sufficiently large datasets are accumulated," they wrote.