NEW YORK — A model combining measures of tumor mutation burden and interferon-ɣ-associated gene expression can predict immunotherapy response in melanoma patients, a new study has found.
For their analysis, a team led by researchers at the University of Sydney examined a number of omics factors that could contribute to treatment response or resistance among advanced melanoma patients receiving immunotherapy.
As they reported Thursday in Cancer Cell, they found that tumor mutational burden, neoantigen load, IFNɣ-associated genes, and T cells in the tumor microenvironment were associated with therapy response, but no specific gene mutation was. A model combining two of these factors — TMB and IFNɣ-associated gene expression — could predict immunotherapy response with high sensitivity. Still, their findings indicated that resistance to checkpoint inhibitor therapy is heterogenous.
"This simply means we have to understand the range of heterogeneity," senior author Georgina Long, co-medical director of the Melanoma Institute Australia and professor at the University of Sydney, said in an email. "For example, there may be some resistance mechanisms that may be somewhat common, and many that are very uncommon. Ultimately, we will need to match a therapy to an individual patient’s driving resistance mechanism."
For their paper, she and her colleagues conducted a multiomic analysis — encompassing whole-genome sequencing, RNA-seq, methylome profiling, and immunohistochemistry analysis — of pretreatment tumor biopsies from 77 patients with advanced cutaneous melanoma and matched germline samples. The patients were later treated with either the anti-programmed death-1 (PD1) therapies nivolumab (Bristol Myers Squibb's Opdivo) or pembrolizumab (Merck's Keytruda) alone or with a PD1 therapy plus ipilimumab (Bristol Myers Squibb's Yervoy), an anti-CTLA-4 therapy.
They flagged a number of factors associated with treatment response. For instance, higher tumor mutation burden and high neoantigen load but low structural variant burden were associated with a good treatment response. Chromothripsis — in line with the structural variant burden finding — was more common among poor responders to treatment.
Additionally, a six-gene IFNɣ signature was more highly expressed among good responders, and, at the same time, the tumor immune microenvironments of responders harbored a higher portion of M1 macrophages and CD8+ T cells.
The researchers were unable to identify any particular gene mutations associated with treatment response, however, despite previous studies implicating certain somatic mutations in poor response to immunotherapy. They suggested that these genes and related mechanisms could have a role in individual patients' treatment resistance but might not be prevalent overall, a finding that hints at heterogeneous treatment resistance mechanisms.
Using the treatment response factors they identified, the researchers developed predictive regression models. Their best model included TMB and IFNɣ-6 and had 89 percent sensitivity, though only 53 percent specificity, in predicting treatment response. In an independent cohort, the model had a sensitivity of 80 percent and a specificity of 59 percent. Long noted that the model is not yet ready for routine use in the clinic.
The model's limited ability to predict treatment resistance could be attributed to the heterogeneity of mechanisms involved, the researchers wrote. They further examined samples from the discordant patients — people predicted to be highly responsive to treatment that were not and vice versa — to try to tease out any biological mechanism for these outliers. But other than a JAK3 loss-of-function mutation, they reported no obvious mechanisms contributing to their outlier status.
Long added that she and her colleagues plan to further investigate patients who did not respond as predicted. "Bulk tissue and patient analysis will not be the way to ultimately overcome resistance," she noted. "We need to get specific and explore/analyze resistance patient by patient."