It has been recognized for more than a century that the immune system possesses an ability to recognize cancer cells as foreign, despite their origins as transformed native cells, and to subsequently destroy them.1 Advances in molecular biology have created novel methods to augment the immune system’s ability to recognize cancer, leading to many treatments currently available for clinical use.2 Collectively, these treatment methods are referred to as immunotherapy.

James Allison and Tasuku Honju recently shared the Nobel Prize for characterizing the ‘immune checkpoint’ molecular interactions of CTLA4 and PD-1/PD-L1, leading to the development of a specific type of immunotherapy.3 Monoclonal antibodies targeting immune checkpoint signaling pathways have become a widely used therapeutic strategy. As of 2019, there are multiple Food and Drug Administration (FDA)approved therapeutics targeting checkpoint inhibitor associated mechanisms and others in clinical trials.4 Despite their relative novelty, checkpoint inhibitors have quickly gained clinical popularity because they are efficacious in multiple cancer types with a favorable safety profile.

Immune checkpoint inhibitors block a tumor’s molecular ability to mask itself from the immune system, thereby exposing tumor cells to the cytotoxic effects of immune effector cells.5 As tumor cells ‘evolve’ from normal cells they consequently take on characteristics that allow the immune system to recognize them as foreign. Under selection from constant immune surveillance, individual tumor clones express checkpoint molecules that act as a strong ‘normal’ signal and thus mask the tumor from immune surveillance. Checkpoint inhibitors disrupt these masking signals.

Not all tumors evade the immune system through identical molecular mechanisms. Heterogeneous mechanisms of immune evasion result in clinical observations that checkpoint inhibitors are not efficacious in all tumor types or in all patients with a particular tumor type. Thus, several biomarkers have been developed in an effort to identify those patients likely to have a clinically meaningful response to checkpoint inhibitor therapy.6

Tumor mutation burden (TMB) is a biomarker with significant recent interest.6 It is derived from analysis of next-generation sequencing (NGS) of tumors and defined as the total number of somatic coding variants observed in a tumor divided by the amount of coding sequence acquired in mega-bases. Many recent retrospective studies have shown that higher TMB values are correlated with improved response rates and survival times with immune checkpoint inhibitor treatment (table 1). Of note, reported TMB calculation methods differ in panel composition and inclusion of all coding variants (including synonymous variants).

Despite the increased clinical utilization of TMB as a biomarker, methods for calculating TMB are not currently harmonized between laboratories (although efforts at harmonization are underway).7 Molecular characterization of tumors is a complex and resource intensive endeavor, making method harmonization a daunting undertaking. However, the impact of changing the fundamental parameters of variant inclusion (the numerator of TMB) and sequencing area (the denominator of TMB) can be tested by in silico analyses of large, publicly available data sets such as The Cancer Genome Atlas (TCGA). Here, we report in silico simulations of six molecular profiling products, including our institution’s in-house NGS solid tumor molecular profiling method. Additionally, we correlate simulation based estimates of TMB with survival data obtained in the absence of checkpoint inhibitor therapy to observe the correlation between survival and TMB.

Molecular characterization of tumors has quickly become standard of care in oncology. Recent survey data indicates that 75% of oncologists across practice types and sizes routinely use NGS data to guide patient care decisions.19 Many commercially available and laboratory-developed NGS based tumor profiling assays also include an evaluation of TMB.

TMB measurement requires multiple steps starting with DNA extraction from tissue, followed by sequencing and alignment, variant identification and informatics-based TMB calculations. Given the complexity and cost of TMB determination, harmonization between laboratories is challenging. A recent review of proficiency testing results from the College of American Pathologists for 111 laboratories indicates that NGS-based testing reliably identifies variants across multiple laboratories in more than 98% of samples tested.20 Thus, identification of individual variants is unlikely to cause significant differences in TMB determinations between laboratories. Instead, discordance between TMB measurement and interpretation is likely to be secondary to differences in variant inclusion (ie, inclusion vs exclusion of synonymous variants), gene panel size and TMB cut-offs used for clinical decision making. The experiment described here addresses the impact of panel size and variant inclusion criteria by calculating TMB from a controlled set of variant lists using differing informatics methods. It should be noted that sample preparation and sequencing methods designed for high sensitivity and specificity in most clinical labs today are likely different from the research methods of the TCGA.

TMB is an aggregate measure of variants identified from sequencing nucleic acids obtained from tumor bulk consisting of both tumor cells and non-neoplastic tumor-associated cells. The current hypothesis that TMB is a surrogate marker for tumor neoantigen formation is widely accepted and consistent with the correlation between high TMB and response to checkpoint inhibitor therapy.21 Given the expectation that TMB is correlated with amino acid changes at the protein level of expressed genes, synonymous mutations should not be included in TMB assessments. Our data indicate that the inclusion of synonymous variants in the calculation of TMB from WES creates a linear bias (figure 2A,B). However, including synonymous variants in TMB derived from gene panels, such as the FoundationOne CDx assay, does not create a discordance between similarly sized panels that exclude synonymous variants (figure 3).

WES is not currently used for routine tumor genome profiling due to cost and clinical utility considerations. Thus, selective gene panels are commonly used for routine patient care. To date, only limited efforts to perform simultaneous WES and smaller panel TMB assessments have been performed (n=29) and results were not definitive (linear regression R²=0.75).22 Previously published methods of estimating the impact of panel size and content on TMB have used stochastic filtering methods where random exome sampling of similar size to the gene panel was performed to assess the correlation between WES TMB and panel TMB.23–25 In contrast, the methods used here for panel TMB estimation used only those genes contained in commonly used panels.

The correlation between panel based and WES TMB values is linear. Thus, in our data set, we see a preservation of the clinical correlation between TMB derived from WES and OS when method-specific quartile values are used to segment patients. TMB values based on any of the panels investigated in this study provide an equivalent estimate of the mutation burden of a tumor; however, decisions based on an individual patient’s TMB value must take into account the source of the data used to calculate the patient’s TMB (eg, UCSD STMP) and the population data source used to inform the decision (WES TMB on a clinical trial population). Reassuringly, the correlation between panel-based TMB values showed slopes of 0.9–1.1 (figure 3D), indicating that the absolute TMB value is comparable between panels.

As with most laboratory values, TMB is often reported with an associated reference range to guide clinical interpretation. The Clinical Genomics Laboratory at UC San Diego reports the calculated TMB value along with a designation of ‘high’, ‘medium’ and ‘low’ based on published TMB population distributions from glioblastoma multiforme, the most frequently profiled tumor type in the laboratory.26 While the value of TMB is analytically equivalent between the methods evaluated, it is important to note that interpretation cut-offs are dependent on the clinical utilization of TMB (eg, predicting response to immunotherapy vs patient prognosis) and the specific tumor type. The patient cohort in this analysis did not receive immunotherapy, thus, cut-offs to predict response to immunotherapy cannot be derived from our analysis. As additional data iare made available, tumor-specific reporting ranges informing response to checkpoint inhibitor therapy, patient prognosis, or both would be preferable. Additional data collection and analysis is needed to establish these ranges.

Our analysis of the correlation between TMB and OS in the absence of checkpoint inhibitor therapy suggests that TMB may be a confounding factor in response to therapy and survival analysis of checkpoint inhibitor therapy in some cancer types. Other investigations have also shown a correlation between TMB and OS in the absence of checkpoint inhibitor therapy.14 15 Consistent with these findings, a recent analysis of data from a prospective trial comparing checkpoint inhibitor therapy to chemotherapy in TMB high NSCLC showed no OS benefit of checkpoint inhibitor therapy despite higher response rates and longer progression-free survival.27 Additional prospective clinical trials using TMB as a biomarker will continue to clarify its clinical utility.

TMB is hypothesized to be a coarse approximation of neoantigen formation. Its relevance to the underlying molecular processes driving and perpetuating tumorigenesis is not clearly known. Our finding that TMB is consistently higher in WES based TMB compared with panel-based TMB is intriguing in that it suggests that the panels analyzed are sampling portions of the genome with lower than average mutation burden. These findings have also been observed in an independent analysis of the TCGA data set.28 Currently, gene panels are typically designed to identify actionable point mutations, indels or copy number variants and not to assess overall TMB. Further investigation may show that mutations in some genes are more readily tolerated than others because of functional redundancy, decreased neoantigen presentation due to loss of major histocompatibility complex (MHC) expression or combinations of other mechanisms. Novel biomarkers, such as predictions of binding and presentation of neoantigens on specific MHC alleles, may provide superior predictive value for prognosis or response to therapy.29 30 As our understanding of immunoncology continues to increase the utility of TMB and other biomarkers will be clarified.