NEW YORK – Applying a universal cutoff for high tumor mutational burden (TMB) across tumor types may not capture all cancer patients who will respond well to immunotherapy, according to an analysis published today in JAMA Oncology. Â
Debates on the relative utility of a universal TMB cutoff versus tumor-specific cutoffs have intensified among oncologists since the US Food and Drug Administration approved pembrolizumab (Merck's Keytruda) last year for refractory patients with any type of solid tumor characterized by high TMB defined as at least 10 mutations per megabase. The agency simultaneously approved Foundation Medicine's FoundationOne CDx next-generation sequencing panel as a companion diagnostic for identifying patients with high TMB who are eligible for pembrolizumab treatment.Â
The latest study, published by researchers from Memorial Sloan Kettering Cancer Center and the Cleveland Clinic, adds to a growing body of data that suggest that a universal high TMB cutoff may not be as precise in predicting who will respond to immunotherapy as tumor-specific TMB cutoffs.
The TMB-based, tissue-agnostic approval for pembrolizumab was based on a 30 percent objective response rate in a cohort of 102 patients with various cancer types enrolled in the Keynote-158 study, all of whom had TMB of least 10 mutations per megabase. Broken down by cancer type, sample sizes were small; there were only five patients with TMB-high neuroendocrine tumors, for example, and two patients with TMB-high thyroid cancers. Given the limited representation of patients with certain tumor types in Keynote-158, many oncologists weren't convinced that the 10 mutations per megabase TMB cutoff was truly predictive of benefit in all histology-specific settings.Â
"We did this study because initially, when this FDA approval was announced, there was no way of knowing what the universal cutoff of 10 meant," said Luc Morris, associate director of MSKCC's immunogenomics and precision oncology platform and one of the authors on the JAMA Oncology paper. "What were response rates for different types of cancer with [at least 10 mutations per megabase]? These data were not available, so we set out to see how the TMB 10 [mutations per megabase] cutoff performed in different types of cancer."
Cutoffs in the 'wrong place'
To address their question, Morris and colleagues analyzed retrospective data from a cohort of 1,678 patients with 16 different cancer types who had been treated with anti-PD-1 or -PD-L1 monotherapy or combination therapy at MSK. The MSK-IMPACT next-generation sequencing panel was used for tumor profiling. To ensure that the responses weren't being driven by patients with microsatellite instability, which also leads to the accumulation of mutations in the tumor, all of the patients in this analysis had tumors that were microsatellite stable, as determined by the MSIsensor test.
When researchers defined TMB-high according to the definition in pembrolizumab's FDA-approved label, at least 10 mutations per megabase, 416 patients from the initial cohort fell into this category. The percentage of TMB-high patients varied significantly from cancer to cancer, however, ranging from 53 percent of melanoma patients to 3 percent of both mesotheliomas and ovarian cancers; none of the kidney cancer patients were deemed to have high TMB. Overall, at this cutoff, TMB-high status was predictive of higher response rates compared to TMB-low status in 11 out of 16 evaluable cancer types.
When researchers drilled further into patients' responses to immunotherapy by tumor type, they found more evidence that responses for TMB-high tumors were not universally higher than TMB-low tumors in every histology. For example, 20 percent of TMB-high gastric cancer patients responded to immunotherapy versus 31 percent of TMB-low patients. Response rates were also higher for TMB-low tumors versus TMB-high tumors in mesothelioma, and hepatobiliary and pancreatic cancer patients. Kidney cancer patients were not evaluable because none of the tumor samples had more than 10 mutations per megabase. Â
Morris said it was somewhat surprising to see some cancers for which responses to checkpoint inhibitors were actually higher in the TMB-low groups than the high groups, and that the variability speaks to the inadequacy of the universal 10 mutations per megabase cutoff.
"This is mainly because the cutoff of 10 is in the wrong place for those cancer types," he explained, "but also in part because TMB has stronger predictive value in some cancer types and weaker or even no predictive value in other cancer types."
In the next part of their study, Morris and colleagues analyzed response rates based on tumor-specific cutoffs that they established specifically for each cancer based on a statistical method called the Youden Index. With the tumor-specific cutoffs, the response rates were higher in TMB-high versus TMB-low tumors for 14 out of 16 of the cancer types. However, response rates were higher in TMB-low patients than TMB-high patients for mesothelioma and esophageal cancers, which speaks to the need to further refine tumor-specific cutoffs.
Morris and his colleagues emphasized in their paper that these tumor-specific cutoffs that they used should be taken with a grain of salt, because they haven't been validated independently and because some of the numbers of patients with individual cancer types in the study were too small to really home in on the optimal cutoff.
"Although the number of patients in this study was higher than the cohort used for [pembrolizumab's] FDA approval, the numbers were still low for some tumor types, limiting generalizability," they wrote. "Cancer type-specific cutoffs were optimized in this data set and require validation in independent data sets."
Because of the need for further validation, the researchers didn't focus on the specific cutoffs in their paper. Still, Morris pointed out that the cutoff values that best separated responders and non-responders in the analysis differed significantly between cancer types. For example, for pancreatic cancer, 2.3 mutations per megabase was the high TMB cutoff, in kidney cancer it was 5.1 mutations per megabase, while in melanoma it was 20.8 mutations per megabase.
Although the specific cutoffs "didn't work perfectly for all 16 cancer types," Morris said, they "did successfully classify tumors more or less likely to respond in 14 cancer types."
According to Morris, his team will perform additional research to validate these cutoff values using data from patients who undergo MSK-IMPACT testing, and he said it would be feasible for other oncologists testing their patients on Foundation's or other NGS platforms to pursue similar work.
Between science and application
Going forward, Morris said that the research doesn't discount TMB as a biomarker so much as call into question its application.
"The underlying immunology here is sound," he said, explaining that the reason TMB-high tumors tend to respond better to checkpoint inhibitors is that they are more likely to present 'non-self' signals to T cells, and that this "robust phenomenon" is part of what makes TMB a very attractive biomarker.
But as the JAMA Oncology paper shows, "there is a bit of a gulf between the science and the application of a fairly crude, simplistic TMB value to all types of cancer," he said.
For practicing oncologists, this means that while FDA's TMB-based tissue-agnostic approval for pembrolizumab may be based on sound underlying immunology, they should not assume that pembrolizumab is necessarily the best option for all refractory cancer patients with a TMB of 10 mutations per megabase. As Morris further explained, the reasons that different cancers may not respond as well to immunotherapy despite being TMB-high according to the FDA pembrolizumab label may be rooted in a number of variables.
Differences in the ways that tumors process and present neoantigens, patients' tumor microenvironments, levels of immune surveillance or immunosuppressive factors, or immune escape during tumor development can all influence immunotherapy responses. Host factors, such as differing HLA genotypes, may also play into variable responses.
"Tumor mutational burden will likely need to be combined with other biomarkers to more robustly stratify responders and nonresponders," Morris and colleagues wrote in their paper. Some possibilities of biomarkers to pair with TMB could include the HLA genotype, or measurements of T-cell infiltration in the tumor microenvironment, Morris suggested.
In the meantime, amid imperfect and advancing science, oncologists are still disagreeing over whether the FDA's tissue-agnostic approval for pembrolizumab using TMB is truly an advance for cancer patients. On the one hand, the approval makes pembrolizumab an option for more late-stage patients who are otherwise out of options. On the other hand, using a universal cutoff, some patients may miss out on the chance to get treatment or receive an expensive therapy when they're unlikely to derive much benefit.
"As physicians, we often favor having access to more drugs and more therapies for our patients," Morris said. "But we often do not consider other factors as often, such as the comparative effectiveness of different therapies, and we rarely think about complex considerations such as cost-effectiveness."
The FDA, however, does not consider cost-effectiveness when approving treatments. The list price for a single 200 mg dose of pembrolizumab — which is indicated for TMB-high patients every three weeks — is nearly $10,000. In addition to cost, pembrolizumab comes with its set of side effects and quality-of-life considerations that doctors must also consider before prescribing it to patients. "This is an area that needs further research by experts in health policy," Morris said.