As immune checkpoint inhibitors are approved for more cancers and indications, researchers continue to hunt for biomarkers that can predict which patients will respond to the therapies. One such biomarker, tumor mutation burden, has potential, but requires better standardization across labs and institutions, as well as stronger prospective clinical trial data, before it can be considered for routine clinical use.

Some patients have strong, durable responses to immune checkpoint inhibitors, but many do not, leaving researchers searching for biomarkers to predict response. Tumor mutation burden (TMB) is one possibility. Researchers have found that patients with a higher TMB have more somatic mutations in their tumor cells generating neoantigens that make their tumors more responsive to the therapies. However, clinical implementation of TMB has proven complicated.

“There are several steps that must be done before we validate TMB,” says Alfredo Addeo, MD, of the University Hospital of Geneva in Switzerland, who co-authored an article on TMB in non–small cell lung cancer (NSCLC; JAMA Oncol 2019;5:934–5). One such step: standardizing TMB calculation.

Kaushal Parikh, MD, of Mayo Clinic in Rochester, MN, and his team began investigating TMB standardization after noticing that some patients with high TMB did not respond as well as expected to immune checkpoint inhibitors. They calculated TMB in 50 patients with 10 different cancers using what Parikh considers the gold standard: subtracting the number of germline mutations from somatic mutations present in each patient's tumor cells. They then compared these TMB values with TMB values calculated for the same patients using three other germline filtering techniques: two based on publicly available databases of somatic and germline variants and one based on Tempus's proprietary algorithm. Parikh presented the findings at the 2019 American Society of Clinical Oncology (ASCO) Annual Meeting (J Clin Oncol 37, 2019 [suppl; abstr 2621]).

These three filtering methods generally calculated higher TMB values than germline subtraction, Parikh says, “which is why we may not be getting the desired responses in our patients.” Such inconsistencies, he adds, become particularly problematic when using TMB to select patients for clinical trials.

Also at the ASCO meeting, a team presented findings from a study investigating how TMB calculations might be standardized across labs (J Clin Oncol 37, 2019 [suppl; abstr 2624]). The central issue, says study co-author Lisa McShane, PhD, of the NCI, is that many labs calculate TMB based on targeted panels that measure mutations in a subset of genes, rather than whole-exome sequencing that captures mutations across the entire genome. “It's possible that the mutations in an individual tumor might fall outside the particular genes on a panel that is being used by a clinical laboratory,” McShane explains. In the first phase of the study, the researchers found that TMB values calculated on the same tumors varied across 11 gene panels, yet there were strong linear relationships between gene panel and whole-exome TMB values. “That is encouraging because it means that we can use calibration procedures” to adjust TMB measurements to be more comparable, McShane says.


Tumor cells with a higher TMB have more somatic mutations producing neoantigens (left) than those with a lower TMB (right). These neoantigens make tumor cells more sensitive to immune checkpoint inhibitors.

In the second phase, the researchers used 10 well-characterized tumor cell lines—eight lung and two breast—to develop reference standards that were sent to 15 labs using gene panels to calculate TMB. Similar to the first phase, TMB values varied across labs, with greater variation at higher TMB levels. However, the researchers also saw strong correlations between TMB values calculated by labs and those determined through whole-exome sequencing, with eight labs consistently over- or underestimating TMB.

“We can't tell labs to change their assay procedures, but what we can do is to help them to better understand how their assessments may systematically differ from assessments of another lab” by developing reference standards for comparison, McShane says. “I don't think it's ever going to be 100% perfect, but if we could … reduce some of the variability, I think that would be really beneficial to patients.”

Parikh and McShane are also curious about how various methods of calculating TMB equate to patient responses, and whether different TMB cutoffs are needed for different cancers—a notion supported by a recent study (Nat Genet 2019;51:202–6). “We don't want to just harp on the methods of calculating TMB,” Parikh says, but also establish the clinical utility of these TMB values.

That's why Addeo stresses the need for prospective clinical data. “The impression we have about TMB very often is that it is a very simple biomarker, that the cutoffs have been well established, that it has been confirmed as a biomarker for prediction of improving patient outcomes, and it should be utilized in real life,” he says. “This is not the case.”

In NSCLC, for example, most trials assessing TMB have been retrospective or examined progression-free survival and objective response rate rather than overall survival (OS), and have used TMB cutoffs ranging from 10 to 20 mutations per megabase. Addeo wants to see prospective clinical trials in which patients are randomized based on TMB levels and treated with immune checkpoint inhibitors or standard therapies, allowing researchers to make OS comparisons based on TMB. Such trials, he says, are needed to validate TMB as a biomarker and establish meaningful TMB cutoffs.

Despite these challenges, Addeo sees potential in using TMB as a biomarker, particularly in NSCLC. “I think TMB remains a fascinating concept, and it's certainly something to further pursue, but we need to take the right steps,” he says. “Otherwise there's a serious risk we might not explore it properly, and therefore we will never use it properly.” –Catherine Caruso

For more news on cancer research, visit Cancer Discovery online at