Tumor Mutational Burden (TMB): From Research to Clinical Reality, 2027 Update

Tumor Mutational Burden (TMB): From Research to Clinical Reality, 2027 Update

Tumor mutational burden (TMB), also called tumor mutation burden or tumor mutational load (tumour mutational burden in UK spelling), gave immunotherapy something it badly wanted: a number, a threshold and a tumour-agnostic label. In June 2020 the FDA approved pembrolizumab for TMB-high solid tumours at 10 mutations per megabase (mut/Mb), with FoundationOne CDx as the companion diagnostic. Six years on, TMB testing is routine in comprehensive genomic profiling reports, but the evidence on what a TMB-high result actually means has become more detailed and, in places, more sceptical.

This refresh of our earlier article sets out what TMB measures, how the 10 mut/Mb cut-off came about, where it predicts benefit and where it does not, what harmonisation work has achieved, and what trial and diagnostics teams should do with TMB in 2027.

Key takeaways

• TMB is the number of somatic coding mutations per megabase of sequenced genome. Whole exome sequencing (WES) is the reference; clinical labs mostly estimate it from large targeted panels.
• The FDA’s June 2020 accelerated approval of pembrolizumab for TMB-H (≥10 mut/Mb) solid tumours rested on 102 patients in KEYNOTE-158, with an objective response rate of 29%.
• Large analyses since then (McGrail 2021, Valero 2021) show TMB-H predicts checkpoint response in some cancer types, such as lung, bladder and melanoma, but not in others, such as breast, prostate and glioma.
• Panel TMB values differ between assays. The Friends of Cancer Research harmonisation project and the 2024 AMP/CAP/SITC recommendations exist to make numbers comparable, but a single universal threshold remains a regulatory convenience rather than a biological constant.
• Blood TMB has not yet earned a place in practice: the randomised BFAST cohort C missed its primary endpoint.
• For trial teams, TMB is most useful as a prespecified, tumour-type-specific stratifier read alongside MSI, PD-L1 and immune context, not as a stand-alone selection tool.

What changed since our earlier version

• Added the tumour-type evidence from McGrail et al. and Valero et al. that underpins today’s debate, and the 2024 real-world analysis of more than 70,000 microsatellite-stable patients.
• Added the Friends of Cancer Research harmonisation phases with their publications, and the 2024 AMP/CAP/SITC consensus on TMB assay validation and reporting.
• Replaced unverified blood TMB figures with the published POPLAR/OAK derivation and the BFAST cohort C randomised result.
• Added a guideline section (FDA label, ESMO Precision Medicine Working Group) and a practical checklist for trial and diagnostics teams.
• Corrected the claim that TMB-H was the first tumour-agnostic genomic approval: MSI-H/dMMR (2017) and NTRK fusions (2018) came first.

TMB analysis: what it measures and how it is calculated

The biological rationale is simple. Somatic mutations in coding regions can produce altered peptides, some of which are presented on HLA class I molecules as neoantigens. More mutations should mean more chances of a neoantigen that T cells recognise, and therefore a better chance that checkpoint blockade unleashes a useful response.

TMB is reported as mutations per megabase. Labs differ on what counts:

• Variant types: some count only non-synonymous single-nucleotide variants; others include synonymous variants and short indels to reduce sampling noise on smaller panels.
• Germline filtering: tumour-only panels must remove germline variants computationally, which is harder in under-represented ancestries and can inflate TMB.
• Driver and hotspot exclusion: known cancer hotspots are often removed because panels are enriched for cancer genes.
• Tumour purity and depth: low cellularity lowers variant allele fractions and can push a true TMB-high tumour below the cut-off.

How TMB is tested: panels, whole exome and blood

A tumor mutational burden test today is almost always a next-generation sequencing (NGS) test. There are three routes, and they do not give identical numbers:

• Whole exome sequencing (WES) reads roughly the whole protein-coding genome, usually with a matched normal sample to remove germline variants. It is the reference method behind most early checkpoint inhibitor studies, but cost and turnaround keep it out of most routine care.
• Large targeted NGS panels sequence a few hundred cancer genes and extrapolate a genome-wide TMB. This is how most patients are tested, because the same run also reports actionable alterations and MSI. FoundationOne CDx is the FDA-approved companion diagnostic for TMB-H (≥10 mut/Mb) with pembrolizumab.
• Blood (ctDNA) panels estimate TMB from circulating tumour DNA when tissue is scarce. Results depend on tumour fraction, and no blood TMB cut-off is validated for treatment decisions (see the blood TMB section below).

Panel size matters. The smaller the sequenced territory, the fewer mutations are counted and the wider the statistical uncertainty around a value near 10 mut/Mb. Variant calling choices (which variant types are counted, how germline variants and hotspots are filtered, and the minimum allele fraction) shift the number further. Sample type matters too: formalin-fixed tissue with low tumour cellularity can under-call mutations, while fixation artefacts can add false ones if they are not filtered.

TMB NGS results from targeted panels are calibrated against whole exome sequencing, which is why the same tumour can score differently on two assays:

How to interpret a TMB score

A TMB result is reported as mutations per megabase (mut/Mb) of sequenced DNA. Reading it well means asking four questions.

1. Which threshold applies? The only regulatory cut-off is TMB-H, 10 or more mut/Mb, set by the June 2020 accelerated approval of pembrolizumab for previously treated solid tumours and measured with FoundationOne CDx. Some reports also label results as low or intermediate, but those bands are assay-specific, not regulatory.
2. Which assay produced it? Values from different panels are not automatically interchangeable. A score of 10 on one assay may correspond to a different value on another, which is why the harmonisation work covered below matters.
3. How close is it to the cut-off? Results just above or below 10 mut/Mb carry the most sampling uncertainty, especially on smaller panels or low-purity samples. A borderline value is a reason for caution, not a clean yes or no.
4. Which tumour type is it? TMB distributions differ widely. Melanoma, lung and bladder cancers often have high TMB, while many breast and prostate cancers sit low, and the predictive value of TMB-H also varies by tumour type. Check MSI or mismatch repair status too: MSI-H/dMMR tumours are usually TMB-high and have their own approvals.

In short, a TMB score is a calibrated estimate from a specific assay, best read alongside tumour type, MSI and PD-L1 rather than as a stand-alone answer.

From research to FDA approval

Early exome studies in melanoma and non-small cell lung cancer (NSCLC) linked higher mutation load with checkpoint response, which fits the high TMB driven by UV light and tobacco in those cancers. A large MSK-IMPACT analysis by Samstein et al. in Nature Genetics (2019) then showed higher TMB associated with better survival after checkpoint inhibitors across many cancer types, though the best cut-off varied by histology.

On 16 June 2020 the FDA granted accelerated approval to pembrolizumab for adults and children with unresectable or metastatic TMB-H (≥10 mut/Mb) solid tumours, determined by an FDA-approved test, that have progressed after prior treatment and have no satisfactory alternative options. FoundationOne CDx was approved as the companion diagnostic.

The evidence came from a prospectively planned retrospective analysis of ten cohorts of the single-arm KEYNOTE-158 trial. Among 102 TMB-H patients (13% of enrolled patients), the objective response rate was 29%, including 4% complete responses; 57% of responders had responses lasting 12 months or more. The FDA notes that safety and effectiveness have not been established in paediatric patients with TMB-H central nervous system cancers. Importantly, several common tumour types, including breast, prostate and colorectal cancer, were not represented in the cohorts behind the approval.

This was the third tumour-agnostic genomic approval, after MSI-H/dMMR in 2017 and NTRK fusions in 2018, and it was the first to depend on a continuous number with a cut-off rather than a qualitative alteration.

Not every sponsor got there. Bristol Myers Squibb withdrew its US application for first-line nivolumab plus ipilimumab in TMB ≥10 NSCLC in January 2019, after discussions with the FDA about how to interpret overall survival by TMB and PD-L1 in CheckMate -227.

The debate since 2020: one threshold, many tumours

Two papers published in 2021 shaped the debate that followed.

McGrail et al., in Annals of Oncology, used more than 10,000 TCGA tumours plus checkpoint outcome data (1,551 patients for response, 1,936 for survival). In cancer types where CD8 T-cell infiltration tracked neoantigen load, such as lung, bladder and melanoma, TMB-H tumours had an objective response rate of 39.8%, with an odds ratio of 4.1 against TMB-low tumours. In cancer types without that relationship, including breast, prostate and brain cancers, TMB-H did not predict response. The authors argued for cancer-type-specific work before broad use.

Valero et al., in JAMA Oncology, studied 1,678 patients treated with checkpoint inhibitors at Memorial Sloan Kettering. Overall, 41% of TMB-H patients responded against 21% of TMB-low patients, but TMB-H was not associated with higher response rates in 5 of 16 cancer types. The share of tumours above 10 mut/Mb ranged from 0% in kidney cancer to 53% in melanoma. The authors concluded that a universal threshold had limited predictive value.

Real-world data have added weight on both sides. A 2025 analysis led by Foundation Medicine and UC Davis investigators in the Journal for ImmunoTherapy of Cancer looked at more than 8,000 patients across 24 cancer types on single-agent checkpoint inhibitors and concluded that 10 mut/Mb on FoundationOne CDx remains a reasonable pan-tumour cut-off for that assay. Critics noted it relies on one assay. Meanwhile a 2024 analysis of 70,698 microsatellite-stable patients in Nature Cancer (Muquith et al.) found that the association between TMB and survival depended on tissue and treatment, supporting tissue-specific thresholds.

Where the signal is weak or absent

Harmonisation: making the number comparable

Because the 10 mut/Mb cut-off was set on one assay, anyone using a different panel has to show their number means the same thing. Two bodies of work matter.

Friends of Cancer Research TMB Harmonization Project

• Phase I (in silico): diagnostic partners calculated panel TMB from the exome subset each panel covers, using TCGA data from 32 cancer types. Panel TMB correlated with exome TMB, but calculation method and panel size affected agreement. Published by Merino et al. in the Journal for ImmunoTherapy of Cancer (2020).
• Phase II (empirical): 16 laboratories tested shared clinical samples and reference materials against WES. Results varied between panels, and the project produced a publicly available calibration approach to align panel values to an exome scale. Published in Annals of Oncology in 2021, as announced by Friends of Cancer Research.

Earlier interim findings showed that panels tended to overestimate TMB at low values and underestimate it at high values, which matters most near a decision threshold.

AMP/CAP/SITC consensus (2024)

In June 2024 the Association for Molecular Pathology, the College of American Pathologists and the Society for Immunotherapy of Cancer published joint recommendations for TMB assay validation and reporting in the Journal of Molecular Diagnostics. They cover pre-analytical, analytical and post-analytical factors, and ask labs to describe their methods in enough detail for results to be compared across assays.

Blood TMB: promising, but not proven

Blood TMB (bTMB) uses ctDNA panels and avoids the tissue problem. Gandara et al. derived a cut-off of 16 or more mutations in POPLAR and confirmed it in OAK, where bTMB-high patients had longer progression-free survival on atezolizumab than docetaxel (HR 0.65) in a retrospective analysis published in Nature Medicine (2018).

The prospective test was cohort C of the phase 3 BFAST trial, which randomised 471 patients with NSCLC and bTMB of 10 or more to atezolizumab or platinum chemotherapy. In the primary population (bTMB ≥16), investigator-assessed progression-free survival did not reach statistical significance (HR 0.77, 95% CI 0.59–1.00, P = 0.053), as reported in Nature Medicine (2022). bTMB therefore remains a research and exploratory tool. Tissue cut-offs cannot be transferred to blood, and low tumour fraction can make samples uninformative. For more on ctDNA in IO trials, see our article on ctDNA and MRD in immunotherapy trials.

TMB with other markers

TMB counts mutations. It does not measure whether those mutations are presented, recognised or acted on. Most of the newer work combines TMB with markers that fill those gaps.

• MSI and MMR status: MSI-H/dMMR tumours are almost always TMB-H and respond well, and they have their own approvals. In practice, read TMB after MSI status, and pay particular attention to TMB-H tumours that are microsatellite stable.
• PD-L1: PD-L1 expression and TMB are largely independent, and each adds information. Our guide to PD-L1 as a predictive biomarker covers the assay side of that story.
• HLA: HLA class I heterozygosity and loss of heterozygosity affect how many neoantigens can be presented, so the same TMB can mean different immunogenicity in different patients.
• Clonality: clonal neoantigens are more likely to drive durable responses than subclonal ones. A pan-cancer meta-analysis by Litchfield et al. (Cell, 2021) found clonal TMB among the strongest predictors of checkpoint response.
• Mutational signatures and immune context: UV, tobacco, APOBEC and mismatch repair signatures, and T-cell-inflamed gene expression or CD8 density, help explain why TMB works in some histologies and not others.

Where TMB stands in guidelines today

• FDA label: pembrolizumab remains indicated for previously treated TMB-H (≥10 mut/Mb) solid tumours with no satisfactory alternatives, determined by an FDA-approved test.
• ESMO Precision Medicine Working Group: its 2020 recommendations advised TMB testing in cervical cancer, well and moderately differentiated neuroendocrine tumours, and salivary, thyroid and vulvar cancers, where TMB-H predicted pembrolizumab response in KEYNOTE-158, and stopped short of recommending it more broadly. The group updated its NGS recommendations in 2024.
• NCCN: several disease-specific NCCN guidelines include TMB-H as a basis for pembrolizumab, mainly in later lines. Check the current version for each tumour type, as positions differ by disease.

In short, TMB is accepted as a tumour-agnostic option for previously treated patients and as one input among several. It is not a first-line selection marker in any major tumour type.

Practical guidance for trial and diagnostics teams

• Prespecify, do not mine. If TMB matters to your hypothesis, define the assay, cut-off and analysis population in the protocol. Post-hoc TMB subgroups have a poor record. Our guide on how to validate an IO biomarker in a clinical trial sets out the steps.
• Analyse by histology. Report results within tumour types and test cut-offs per histology, not only pan-tumour.
• Lock the assay early. Choose one panel and pipeline, document variant types counted and germline filtering, and calibrate to WES if you plan to bridge to another assay.
• Record purity and ancestry. Capture tumour content and consider ancestry-related germline filtering bias, especially in tumour-only testing.
• Collect the companions. Bank material for MSI, PD-L1, HLA typing and immune gene expression so that composite models can be tested later. Designs are covered in our article on biomarker-driven clinical trials.
• Decide the regulatory role early. A TMB result that selects patients needs a companion diagnostic; one that informs but does not restrict use may fit a complementary diagnostic. See companion vs complementary diagnostics.
• Treat blood TMB as exploratory unless you are prepared to run the prospective validation BFAST showed is needed.

Discuss predictive biomarkers at ImmunoMark London

TMB is a case study in what happens when a biomarker moves faster than its evidence base, and in how harmonisation, regulatory and clinical teams have had to catch up. Those questions sit squarely in Stream 2: Biomarkers & Companion Diagnostics at the biomarker and immuno-oncology conference in London on 3–4 March 2027, the second annual UK edition, with a target of 400+ attendees.

The London 2027 Stream 2 agenda includes a keynote on “Emerging Predictive & Prognostic Biomarkers in IO”, “Keynote : Clinical Utility of Biomarkers” from Espen Walker, Head of Global Medical Diagnostics, AstraZeneca, and a “Panel: Regulatory Harmonisation for Biomarker Qualification”. The format combines keynotes, panels, case studies, pre-arranged 1:1 meetings and a Day 1 drinks reception. The series also runs in San Diego on 10–11 February 2027 and in Boston in June 2027.

Register for ImmunoMark London 2027

If your company wants to meet this audience directly, whether you run TMB and CGP testing, supply sequencing or bioinformatics platforms, or offer biomarker CRO services, see the London 2027 sponsorship opportunities or our overview of sponsoring a biomarker conference.

FAQs

What is tumor mutational burden?

Tumor mutational burden (TMB) is the number of somatic mutations in the coding regions of a tumour genome, reported as mutations per megabase (mut/Mb). Whole exome sequencing is the reference method, but most clinical tests estimate TMB from large targeted gene panels. More mutations can mean more neoantigens for T cells to recognise, which is why TMB is studied as a predictor of response to immune checkpoint inhibitors.

What does TMB high mean?

For the FDA-approved use of pembrolizumab, TMB high (TMB-H) means 10 or more mutations per megabase measured by an FDA-approved test, which at approval was FoundationOne CDx. The approval covers previously treated unresectable or metastatic solid tumours with no satisfactory alternative options. Other assays report TMB on their own scales, so a value of 10 on one panel may not equal 10 on another.

How is a TMB test done?

A TMB test is a next-generation sequencing test on tumour tissue, usually a formalin-fixed biopsy or resection sample. Most labs use a large targeted gene panel, such as FoundationOne CDx, which also reports actionable mutations and MSI in the same run. Whole exome sequencing is the research reference, and blood-based ctDNA panels can estimate TMB when tissue is not available, though blood TMB is not validated for treatment decisions.

What is a high TMB score?

For the FDA-approved use of pembrolizumab, a high TMB score is 10 or more mutations per megabase measured with an FDA-approved test, FoundationOne CDx. Most tumours score well below that. Some reports add low or intermediate bands, but these are specific to the assay. Values close to 10 mut/Mb carry the most uncertainty, so the assay, sample quality and tumour type should be considered when reading a borderline result.

Why do different TMB tests give different results?

Panels sequence different amounts of DNA, count different variant types and filter germline variants and hotspots in different ways, so the same tumour can produce different TMB values on two assays. Smaller panels have more sampling error, and low tumour purity can lower the count. Harmonisation work by Friends of Cancer Research and the 2024 AMP/CAP/SITC recommendations aim to make results more comparable.

Is the TMB cut-off of 10 mut/Mb valid for all cancers?

Not equally. Studies by McGrail et al. and Valero et al. in 2021 found TMB-H predicted checkpoint response in cancers such as lung, bladder and melanoma, but not in breast, prostate or brain cancers, and the share of tumours above 10 mut/Mb varies widely by cancer type. A 2025 real-world analysis on FoundationOne CDx supported 10 mut/Mb as a reasonable cut-off for that assay, but many researchers favour tumour-type-specific thresholds.

Can TMB be measured from blood?

Yes, blood TMB can be estimated from circulating tumour DNA, but it is not validated for treatment decisions. In the phase 3 BFAST cohort C trial in NSCLC, atezolizumab did not significantly improve progression-free survival over chemotherapy in patients with blood TMB of 16 or more. Tissue cut-offs do not transfer directly to blood, and samples with low tumour fraction can be uninformative.

How should TMB be used alongside PD-L1 and MSI?

Check MSI or mismatch repair status first, because MSI-H/dMMR tumours are almost always TMB-high and have their own approvals. PD-L1 and TMB are largely independent and each adds information. Trial teams increasingly combine TMB with PD-L1, HLA status, clonality and immune gene expression rather than using it as a stand-alone selection marker.