Increasingly, oncologists are turning to tumor mutation burden (TMB) as a biomarker of response to immune checkpoint inhibitors. However, recent research suggests that TMB needs to be further refined before it can be adopted broadly across tumor types and patient populations.

Increasingly, oncologists are turning to tumor mutation burden (TMB) as a biomarker of response to immune checkpoint inhibitors (ICI). In fact, based on the KEYNOTE-158 trial, the FDA last year granted accelerated approval to pembrolizumab (Keytruda; Merck) for inoperable or metastatic solid cancers identified as TMB-high (TMB-H), with at least 10 mutations per megabase (mut/Mb). However, recent research suggests that TMB needs further refining before it can be adopted broadly across tumor types and patient populations.

The concept is quite simple: Patients with high TMB have more neoantigen-generating somatic mutations, making their tumors more responsive to ICIs. In KEYNOTE-158, 102 TMB-H patients across 10 tumor-type cohorts had an overall response rate (ORR) of 29%.

Yet the trial cohorts were relatively limited in size and scope, noted Daniel McGrail, PhD, of The University of Texas MD Anderson Cancer Center in Houston. He and his team performed a retrospective TMB analysis that included more patients and tumor types (Ann Onc 2021;32:661–72). When neoantigen load positively correlated with CD8 T-cell levels—such as in bladder cancer and melanoma—patients with TMB-H tumors had an ORR of 39.8%, compared with 15.2% in their TMB-low (TMB-L) counterparts. In contrast, in breast, prostate, and brain cancers, where there was no such correlation, the ORR was 15.3% among TMB-H patients, and 23.4% in corresponding TMB-L patients.

“While TMB does appear to predict response in a subset of cancers, this is not necessarily universally true,” McGrail said, and these results “definitely urge caution” in interpreting TMB. “We want to make sure we utilize this approval correctly,” he added, by treating patients most likely to benefit, but avoiding ICIs in potentially less effective contexts.

One such context is glioblastoma: Counterintuitively, patients with recurrent disease and very low TMB (<0.6 mut/Mb) live longer after receiving ICIs than TMB-H patients (Nat Comm 2021;12:352). Further work suggests that patients with the lowest TMB have the most active tumor microenvironment, “so it's almost the complete opposite to some other tumors,” explained senior author David Ashley, MBBS, PhD, of Duke University School of Medicine in Durham, NC. “There are a bunch of tumors where the rules don't apply like they might in colorectal cancer or melanoma,” he added and thus “there needs to be a rethink” on the FDA approval—one that includes full prospective trials for each tumor type.

Such trials not only may reveal cancers for which TMB should be used, but also may help identify TMB cutoffs for different malignancies. “It's possible in some of these cancers a higher threshold might still be predictive,” McGrail said.

Research at the virtual American Association for Cancer Research Annual Meeting 2021 further explored TMB cutoffs in patients with advanced solid tumors treated with another ICI, atezolizumab (Tecentriq; Roche). Patients with a TMB of at least 16 mut/Mb had an ORR of 38.1%, versus 2.1% in patients with a TMB between 10 and 16 mut/Mb. Although responses clearly increase as TMB goes up, “I do think we have some issue at the low end of the spectrum,” said presenter John Hainesworth, MD, of Sarah Cannon Research Institute in Nashville, TN, noting that there was also variation by tumor type.

graphic

Pembrolizumab is approved for the treatment of solid cancers with a high tumor mutation burden (TMB). However, for a subset of cancers, high TMB doesn't predict a response.

Another key step in refining TMB is standardizing and improving how it is calculated. Most TMB assays—including Foundation Medicine's FoundationOne CDx, approved as a companion diagnostic for pembrolizumab—use targeted oncogene panels rather than sequencing the entire exome. This approach is streamlined, McGrail said, but it can lead to biased results depending on which oncogenes are present in a cancer. Cancer-specific panels may be necessary—a premise explored in a recent study encompassing 14 different malignancies (NPJ Precis Oncol 2021;5:31).

Targeted panels may also fail to factor in other relevant genomic alterations. In mesothelioma, for example, Aaron Mansfield, MD, of the Mayo Clinic in Rochester, MN, and his team established that patients with a TMB of less than 2 mut/Mb still respond to ICIs. Follow-up research revealed that these patients have a high gene fusion burden typically not captured by oncogene panels, but that likely contributes to neoantigen load (J Thorac Oncol 2019;14:276–87).

In another study, the researchers calculated TMB in patients with multiple myeloma. When they used public germline variant databases, rather than patient-paired germline sequencing, to filter out nonsomatic mutations, TMB estimates were significantly higher—and this inflation was more pronounced in Black patients (NPJ Precis Oncol 2021;5:22). The results suggest that patient-paired germline subtraction is the best way to calculate TMB, Mansfield said—and perhaps calculation method should be factored into TMB cutoffs.

For Ashley, such TMB discrepancies emphasize the need for an “industry-wide benchmark for how the assays are done and validated.” Almost every current assay “will come out with a different median TMB in different tumors,” he added, depending on the calculation method—and in some cases, differences could be substantial.

Ashley also sees a need for more research on TMB's biological underpinnings, which not only may elucidate how the biomarker works in different cancers but could also identify other relevant factors such as race or sex. Further research may also find additional biomarkers, such as microsatellite instability and neutrophil–leukocyte ratio, that could be combined with TMB to improve its predictive power. “TMB is the end result of the immunologic environment,” Ashley said. “It can act as a biomarker in the right circumstances, but you have to look at it contextually.” –Catherine Caruso