Tumor Mutational Burden: clinical utility in immunotherapy

On June 16, 2020, the U.S. Food and Drug Administration granted accelerated tissue-agnostic approval to pembrolizumab for adult and pediatric patients with unresectable or metastatic solid tumors harboring a tumor mutational burden of 10 or more mutations per megabase. The decision established a biomarker-driven route to treatment across several tumor types, but it did not turn TMB into a universal predictor that performs identically in every histology. The clinical meaning of a TMB-high result still depends on tumor biology, assay design, treatment context, and the evidence supporting a particular indication.

The approval rested on data from 102 patients enrolled across 10 disease-specific arms of the phase II KEYNOTE-158 trial. In this population, pembrolizumab produced an overall response rate of 29% (95% confidence interval, 21–39%). Among responders, 57% maintained a response for at least 12 months, while 50% maintained a response for at least 24 months. These figures remain central to the practical interpretation of TMB-high disease. They also explain why the laboratory details matter: once a threshold becomes part of treatment selection, the number reported by the assay is no longer merely descriptive.

Biological Rationale: Neoantigen Presentation and Immune Recognition

Tumor mutational burden estimates the number of nonsynonymous somatic mutations found within a defined genomic territory, usually normalized to one megabase of sequenced coding DNA. In routine practice, that territory is often represented by a targeted next-generation sequencing panel rather than by whole-exome sequencing. The result is expressed as mutations per megabase, or mut/Mb.

The biological premise is straightforward, although the clinical reality is not. A nonsynonymous mutation can alter a protein sequence and create a peptide that was not present in normal tissue. If that peptide is processed, loaded onto a compatible human leukocyte antigen molecule, and displayed on the tumor-cell surface, it may be recognized as foreign by T cells. A tumor with more mutations therefore has a greater statistical opportunity to generate immunogenic neoantigens.

Checkpoint inhibitors do not create this recognition from nothing. Agents such as pembrolizumab block the PD-1 pathway and can release an existing or partially suppressed immune response. The response still depends on whether tumor-derived antigens are presented, whether relevant T-cell clones are present, whether those cells can reach the tumor, and whether additional immune-suppressive mechanisms prevent effective killing.

This is why TMB should be understood as a probabilistic biomarker rather than a direct measurement of immune sensitivity. Mutation density is related to the potential neoantigen repertoire, but it does not tell the laboratory or clinician which mutations produce immunogenic peptides. It does not capture HLA genotype, the efficiency of antigen processing, the clonality of mutations, or the condition of the tumor microenvironment. Two tumors with the same TMB may therefore have different immune landscapes and different responses to checkpoint blockade.

The association between mutation load and immunotherapy response first became especially visible in tumor types such as melanoma, smoking-associated non-small cell lung cancer, and microsatellite instability-high colorectal carcinoma. These diseases can carry substantial numbers of mutations, although the mechanisms producing that mutational burden are not identical. TMB helped translate this broader biological observation into a standardized, reportable variable.

That translation has limits. A high number of mutations can increase the likelihood of immune recognition, but it does not guarantee it. Conversely, a tumor with a lower TMB may still respond when other features — including PD-L1 expression, microsatellite instability, viral antigens, or a favorable immune microenvironment — support checkpoint activity. TMB is best interpreted as one component of an immunotherapy biomarker framework, not as a replacement for pathology, clinical history, or disease-specific evidence.

TMB measures mutational opportunity, not immune response itself. The number becomes clinically useful only when the assay, threshold, tumor context, and treatment evidence are interpreted together.

The 10 mut/Mb Threshold: Clinical Evidence from KEYNOTE-158

The pembrolizumab indication was supported by a prospective-retrospective analysis of the multicohort KEYNOTE-158 study. Patients had advanced solid tumors that had progressed after prior systemic therapy and were enrolled in 10 predefined tumor-type cohorts. The TMB-high subgroup was defined as having at least 10 mutations per megabase according to the FoundationOne CDx hybrid-capture panel.

The reported overall response rate was 29%, with a complete response rate in the low single-digit range and partial responses observed across multiple histologies. The duration of response was clinically important: among patients who responded, 57% continued to respond for at least 12 months and 50% for at least 24 months. Durable responses of this kind are one reason the result attracted attention beyond the tumor types traditionally viewed as highly responsive to checkpoint inhibition.

The regulatory interpretation was tissue-agnostic. That means the indication was framed around a molecular characteristic found across solid tumors rather than restricted to one organ of origin. It does not mean that TMB functions as a histology-independent predictor with equivalent performance in every cancer type. The strength of the association, the baseline probability of response, and the reliability of the threshold can vary by histology and by the biological processes that generated the mutations.

The KEYNOTE-158 findings also should not be read as evidence that responses were evenly distributed across all participating cancers. The study included several tumor cohorts, and the response experience was not a flat or interchangeable pattern across histologies. A tissue-agnostic approval is a regulatory and clinical-use framework; it is not a claim that tumor type has ceased to matter.

The 10 mut/Mb cut-point is similarly easy to overinterpret. It is useful because it creates an operational boundary for a defined treatment indication, but it is not a universal biological switch. A tumor at 9.8 mut/Mb is not biologically categorically different from one at 10.1 mut/Mb. Near the boundary, sampling, sequencing quality, variant filtering, panel size, and tumor purity can all influence the reported value.

The practical question for a pathology laboratory is therefore not simply whether a sample crosses 10 mut/Mb. It is whether the measurement is analytically credible, whether the assay has been validated for the intended use, and whether the result is being interpreted within the clinical setting that supports the treatment decision.

What the threshold does — and does not — establish

A TMB result can support several different kinds of clinical reasoning:

• It can identify a tumor that meets the biomarker definition used in the tissue-agnostic pembrolizumab indication.
• It can contribute to a broader assessment of likely sensitivity to immune checkpoint blockade.
• It can provide a quantitative variable for clinical trials and translational research.
• It can help place other biomarkers, such as microsatellite instability and PD-L1 expression, in context.

It cannot, by itself, establish that a patient will respond, determine how long a response will last, or explain why a tumor with a high value is resistant. Nor does the result remove the need to assess performance status, previous treatment, organ function, disease tempo, immune-related risks, and the approved treatment context.

The threshold is also inseparable from the assay on which the evidence was generated. The same tumor may produce different numerical estimates when tested with panels of different size, gene content, chemistry, coverage, and variant-filtering rules. This is the central reason that TMB reporting requires more than attaching a number to a pathology report.

Regulatory Landscape: FDA Approval and Companion Diagnostics

The tissue-agnostic pembrolizumab indication was accompanied by FDA approval of FoundationOne CDx as a companion diagnostic for tissue-based TMB measurement. FoundationOne CDx uses hybrid capture across a 324-gene panel, providing a defined analytical framework for counting and normalizing qualifying mutations.

This regulatory pairing matters for two reasons. First, the clinical evidence supporting the 10 mut/Mb threshold was generated using a specified assay architecture. Second, the companion diagnostic designation gives laboratories and clinicians a reference point for analytical validation, performance assessment, and reporting.

It does not mean that every clinical laboratory-developed test is categorically excluded from treatment-related decision-making. Alternative assays may be used in appropriate settings, but their results require careful validation and interpretation. A laboratory must be able to explain what its panel measures, how variants are classified, how artifacts are removed, and how its result relates to the evidence behind the treatment indication. Local regulatory requirements, institutional policy, payer rules, and the wording of the relevant drug label may also affect how a result is used.

The distinction between an FDA-approved companion diagnostic and a laboratory-developed assay should therefore be stated precisely. FoundationOne CDx is an FDA-approved companion diagnostic for the relevant tissue-based TMB application. That status does not make every other assay clinically equivalent, but it also does not justify describing the platform as universally mandatory in every setting.

The treatment label supplied the clinical framework, while the laboratory had to supply the analytical reliability. For that reason, a useful report should not present TMB as a context-free number. It should identify the assay or method, the reported value, the interpretation category, and any limitations that could affect confidence in a threshold-adjacent result.

The dosing schedule is not a laboratory characteristic of TMB. It reflects the pembrolizumab label and the cross-tumor nature of the indication. The laboratory’s role is to produce a valid biomarker result; it does not infer treatment suitability from TMB alone.

The same separation applies to reimbursement. A result may be clinically informative without automatically resolving coverage, authorization, or treatment access. Those decisions can depend on the jurisdiction, payer, indication, assay status, and documentation required by the treating institution. Precision medicine works poorly when analytical validity, clinical validity, and administrative eligibility are treated as interchangeable concepts.

Standardization Challenges: Harmonizing Panel-to-Panel Variation

A 10 mut/Mb threshold appears precise, but the measurement behind it contains several sources of uncertainty. Targeted panels differ in the amount of coding sequence covered, the genes included, the sequencing chemistry, the minimum depth of coverage, and the computational pipeline used for variant calling. They may also use different rules for excluding germline variants, synonymous variants, low-allele-frequency calls, and technical artifacts.

Consider two panels applied to the same tumor. One may cover approximately 0.5 megabases of relevant coding sequence, while another covers approximately 1.5 megabases. Their raw mutation counts cannot be compared directly. Each count must be normalized to the callable territory, but the resulting mut/Mb estimates may still differ because a smaller panel samples fewer genomic regions and is more vulnerable to random variation. Panel composition also matters: a panel enriched for genes that are commonly mutated in a particular cancer may behave differently from one designed primarily for broad genomic profiling.

The Friends of Cancer Research TMB Harmonization Project has addressed this problem through a multi-stakeholder effort involving academic centers, assay manufacturers, regulators, and pharmaceutical sponsors. Its work has used standardized cell-line and clinical tumor samples tested across multiple panel designs and chemistries. The purpose is not to make every assay identical, which is unrealistic, but to identify sources of systematic difference and improve comparability.

Three analytical domains are particularly important.

Panel design and callable territory

The larger the genomic territory, the more information an assay has available for estimating mutation density. A smaller panel may still be clinically useful for selected genomic alterations, but its TMB estimate can have wider uncertainty, particularly near a treatment threshold. The list of genes and the definition of callable bases should be documented rather than hidden behind a single final number.

Variant classification and filtering

The mutation count depends on which variants are accepted as qualifying somatic mutations. Pipelines may differ in their handling of synonymous changes, common germline polymorphisms, low-allele-frequency variants, sequencing artifacts, and mutations associated with clonal hematopoiesis. Formalin fixation can introduce damage-related artifacts, while low sequencing depth can make true variants difficult to distinguish from noise.

Filtering is not a minor technical step. It directly changes the numerator in the mut/Mb calculation. Two laboratories can sequence similar material and still report different values if their variant inclusion rules are not aligned.

Tumor purity, ploidy, and sample quality

Tumor purity describes the proportion of nucleated cells in a specimen that are malignant. When the tumor fraction is low, somatic variants may appear at reduced allelic fractions and fall below the assay’s detection threshold. Stromal or inflammatory cells can dilute the signal. Necrosis, limited tissue, DNA degradation, and uneven coverage create additional problems.

Ploidy and copy-number changes can complicate interpretation as well. A mutation present in a subclone may be easier or harder to detect depending on the surrounding genomic architecture. These issues do not simply produce random noise; in some samples, they can systematically lower or distort the estimated TMB.

Reporting uncertainty near the cut-point

A report of 10 mut/Mb can look definitive even when the underlying estimate is sensitive to a small number of variant calls. Laboratories should have a policy for threshold-adjacent results and should communicate limitations clearly. Depending on the assay, that may include an interpretive comment, a quality metric, a tumor-purity estimate, or a recommendation to correlate with the broader clinical and pathological picture.

The aim is not to replace the result with an elaborate disclaimer. It is to prevent false precision. A TMB value is meaningful only when the reader knows how it was generated and whether the specimen met the assay’s quality requirements.

The 10 mut/Mb threshold is a regulatory cut-point derived from a defined assay and clinical cohort. It should not be treated as an assay-independent biological constant.

Emerging Frontiers: Liquid Biopsy and Blood-Based TMB Utility

Tissue is not always available when an immunotherapy decision must be made. Metastatic lesions can be difficult to access, archival tissue may have been consumed by prior testing, and a new biopsy may be unsafe or impractical. Blood-based TMB, derived from cell-free DNA in plasma, offers a minimally invasive alternative to tissue sampling.

The analytical workflow is related to tissue testing but not interchangeable with it. Plasma is collected and processed to isolate cell-free DNA. The DNA is then analyzed with a targeted sequencing panel designed to detect variants that may be present at low allele fractions. Qualifying mutations are counted and normalized to the assay’s covered territory, producing a blood-based estimate of mutational burden.

The difference between the matrices is fundamental. In tissue, the assay measures DNA from a selected lesion or specimen that may contain a variable mixture of tumor and non-tumor cells. In plasma, the signal depends on how much DNA the tumor sheds into circulation. A patient with a substantial tumor burden may still have a low circulating tumor fraction, while a biologically active lesion may shed more DNA than its anatomical size would suggest. The blood result therefore reflects both tumor biology and the ability of the tumor to contribute detectable DNA to the plasma sample.

The clinical evidence for bTMB has been developed most extensively in non-small cell lung cancer. Retrospective analyses of trials evaluating atezolizumab and other checkpoint inhibitors reported associations between elevated blood-based TMB and improved outcomes in selected treatment groups. Those findings supported bTMB as a stratification variable and encouraged further work on blood-based immunotherapy biomarkers.

They did not establish a universal blood-based threshold that can simply be substituted for the tissue-based 10 mut/Mb definition. The FDA has not approved a bTMB cut-point as a tissue-agnostic companion diagnostic for pembrolizumab. Tissue TMB and blood TMB should therefore be treated as related but distinct measurements.

Several analytical issues remain central:

• Plasma collection and processing: delays, inappropriate storage, and insufficient plasma input can reduce the quality of the cell-free DNA signal.
• Tumor fraction: a low fraction of circulating tumor DNA can make true somatic variants difficult to distinguish from background DNA.
• Clonal hematopoiesis: blood-cell clones can release DNA carrying mutations that are not present in the tumor, potentially inflating the apparent mutation burden if they are not recognized.
• Limit of detection: broad panels and deep sequencing improve the chance of detecting low-frequency variants, but they do not eliminate uncertainty when the available DNA mass is small.
• Panel harmonization: different bTMB assays may use different genomic footprints and filtering rules, limiting direct comparison across platforms.

Liquid biopsy is therefore best understood as a parallel measurement architecture, not as a simple replacement for tissue. A blood result may be valuable when tissue is unavailable, but a negative or low bTMB result can reflect limited tumor shedding rather than an absence of clinically relevant tumor mutations. In some cases, tissue remains the more informative matrix; in others, plasma can provide a practical route to molecular information that would otherwise be inaccessible.

Integrating TMB with Other Immunotherapy Biomarkers

TMB rarely operates alone in a modern molecular pathology workflow. PD-L1 expression by immunohistochemistry, microsatellite instability status, mismatch repair deficiency, and other tumor features may contribute to the treatment discussion. These biomarkers overlap in some patients but measure different biological processes.

PD-L1 immunohistochemistry assesses protein expression in a defined cellular context and depends on the antibody, scoring system, specimen type, and tumor setting. Microsatellite instability and mismatch repair deficiency identify a pattern of genomic instability that can produce many mutations and neoantigens, but they are not identical to a global TMB measurement. A tumor can be TMB-high without being microsatellite instability-high, and the reverse relationship can also be clinically relevant depending on the assay and disease.

Combining biomarkers does not automatically produce a better prediction. Each test adds potential information, but each also brings pre-analytical, analytical, and interpretive limitations. The right question is not whether TMB should replace PD-L1 or MSI testing. It is how the results should be positioned within the specific disease and treatment pathway.

For laboratories, this favors integrated reporting rather than isolated numbers. A molecular report can place TMB alongside microsatellite instability or mismatch repair findings when those tests are available, while preserving the distinction between what was directly measured and what is inferred. It should also make clear whether the assay was validated for tissue, plasma, or both.

The phrase “clinical utility” is often used too broadly. In practical terms, checking clinical utility in immunotherapy means asking several linked questions:

1. Analytical validity: Does the assay measure TMB reproducibly in the specimen type being tested?

2. Clinical validity: Is the reported value associated with treatment response in the relevant tumor and clinical setting?

3. Clinical actionability: Does the result correspond to an approved indication, a trial criterion, or another defined treatment pathway?

4. Interpretive context: Are other biomarkers, sample-quality issues, and disease-specific factors likely to change the meaning of the result?

5. Operational reliability: Can the laboratory deliver the result quickly enough and with sufficient documentation to inform care?

This is how to check clinical utility in immunotherapy without reducing the question to a threshold lookup. The number matters, but the chain connecting specimen to sequencing, sequencing to interpretation, and interpretation to treatment matters just as much.

Outlook: Toward Integrated Biomarker Frameworks

TMB has moved from a research measurement to a regulated biomarker used in a defined immunotherapy indication. The pembrolizumab approval demonstrated that a genomic feature shared across different solid tumors can support a tissue-agnostic treatment framework. It did not erase histological differences, eliminate biological uncertainty, or make all TMB assays interchangeable.

The next stage is likely to focus on integration and calibration. Laboratories will continue to work on harmonizing panel-to-panel variation, improving the handling of low-purity specimens, and defining how tissue and blood-based measurements should be interpreted. Bioinformatic pipelines will need to distinguish tumor-derived variants from germline variation, technical artifacts, and mutations arising from clonal hematopoiesis. Reporting systems will need to preserve the quantitative result while making uncertainty visible when it could affect a treatment decision.

There is also an unresolved biological question: not every mutation contributes equally to immune recognition. The clonality of a mutation, its expression, the resulting peptide, HLA binding, antigen processing, and the immune microenvironment all influence whether mutational burden becomes meaningful immune visibility. A future biomarker framework may combine TMB with these features rather than treating mutation count as the endpoint.

For now, the practical position is more disciplined. A TMB-high result can support pembrolizumab selection under the relevant tissue-agnostic indication, but it remains a probabilistic signal whose performance varies across tumor contexts. FoundationOne CDx is an FDA-approved companion diagnostic for the tissue-based application that supported the approval; other assays require appropriate validation and should not be assumed equivalent without evidence. Blood-based TMB may expand access to molecular testing, but it has its own analytical and biological constraints.

The value of TMB lies neither in the number alone nor in the promise of a universal predictor. It lies in a reproducible measurement connected to a defined clinical question. That is the standard pathology laboratories must meet if the biomarker is to remain useful as immunotherapy moves toward increasingly integrated precision-medicine decisions.