Tumor Mutational Burden (TMB): What It Measures, What It Doesn't, and How to Interpret It

By Lociven · NeoantigenLab · July 2026

Your pathology report comes back with a TMB of 14 mut/Mb. The oncologist asks whether this patient is likely to respond to pembrolizumab. What do you actually say?

Tumor mutational burden has become one of the most frequently cited biomarkers in solid tumor immunotherapy — and one of the most frequently misunderstood. This post explains what TMB measures, how it is calculated, what the FDA approval actually covers, and what it cannot tell you.

What TMB measures — and what it doesn't

TMB is a count of somatic mutations per megabase of coding sequence sequenced. That is all it is. It does not measure:

Whether any of those mutations produce peptides that bind HLA
Whether those peptides are actually presented on tumor cell surfaces
Whether the patient has T cells that recognize them
Whether the tumor microenvironment is immunologically accessible

TMB is a proxy. More mutations → more potential neoantigens → higher prior probability that some T cell responses exist → better chance that removing the PD-1/PD-L1 brake does something useful.

The logic holds in aggregate across large cohorts. In individual patients, it breaks down regularly.

How TMB is calculated

The formula is straightforward:

TMB = (number of somatic nonsynonymous mutations) / (Mb of coding sequence covered)

In practice, what counts as a "mutation" varies by platform:

WES-based TMB: Uses the full exome (~30–40 Mb covered). Counts SNVs and indels; typically filters germline variants using a matched normal. Most research studies use this.
Panel-based TMB: Uses targeted gene panels (0.5–2 Mb covered). Foundation Medicine's F1CDx (~1.1 Mb) and MSK-IMPACT (~1.2 Mb) are the most widely used clinical platforms. Panel TMB correlates well with WES TMB but not perfectly — panel size and gene content matter.

Variants of uncertain significance, synonymous mutations, and known germline polymorphisms (dbSNP) are typically excluded. Some pipelines also exclude copy number variants and structural variants. This is why TMB numbers from different labs on the same tumor can differ.

The FDA approval and what the cutoff means

In June 2020, the FDA approved pembrolizumab for adult and pediatric patients with unresectable or metastatic TMB-high solid tumors (≥10 mut/Mb by F1CDx) that have progressed after prior treatment. This was a tumor-agnostic approval — the first of its kind based on a genomic biomarker.

The supporting data came from KEYNOTE-158, a basket trial across 10 tumor types. The overall objective response rate in TMB-high patients was 29%, versus 6% in TMB-low patients.

Several things are worth noting about that cutoff:

10 mut/Mb is not a universal threshold. It was derived from F1CDx data and applies to that platform. The same tumor sequenced on a different panel may come out above or below 10 depending on panel coverage.
The predictive value is tumor-type dependent. In colorectal cancer and glioma, TMB-high status did not predict response to pembrolizumab in KEYNOTE-158. The approval came with a note that "the tumor type or histology should be considered when selecting patients."
MSI-H/dMMR tumors are a separate category. Microsatellite instability-high tumors are almost always TMB-high, but TMB-high does not imply MSI-H. The two biomarkers identify overlapping but distinct patient populations.

TMB distribution by tumor type

Figure 1. Typical TMB range by tumor type. The FDA threshold of 10 mut/Mb is shown as a dashed line. Values are approximate medians from TCGA and clinical sequencing data.

TMB varies dramatically across cancer types. This is not random — it reflects underlying mutational processes:

High TMB (>10 mut/Mb typical): Melanoma (UV damage), lung squamous/adenocarcinoma (tobacco), bladder (APOBEC), colorectal MSI-H (mismatch repair deficiency), endometrial POLE-mutant
Intermediate TMB (1–10 mut/Mb): Most solid tumors — breast, ovarian, pancreatic, prostate
Low TMB (<1 mut/Mb): Pediatric cancers, glioma, thyroid, most hematologic malignancies

Knowing your tumor type's typical TMB range puts a reported value in context. A TMB of 8 mut/Mb in a pancreatic cancer patient is actually high relative to that tumor type; the same value in a melanoma patient is unremarkable.

Where TMB fails as a predictor

Understanding why TMB fails in specific contexts is as important as knowing when it works:

Clonal vs. subclonal mutations. Clonal mutations are present in every tumor cell — neoantigens derived from them are displayed uniformly across the tumor. Subclonal mutations appear in only a fraction of cells. A tumor with many subclonal mutations can have a high TMB but poor immune recognition because no single neoantigen is present on enough cells to drive a dominant T cell response. Clonal TMB (counting only clonal mutations) is a better predictor than total TMB in some datasets.

Immunosuppressive tumor microenvironment. Even high-TMB tumors fail to respond to checkpoint blockade if the tumor microenvironment is cold — low TIL infiltration, high TGF-β signaling, functional exclusion of T cells. TIDE score (Tumor Immune Dysfunction and Exclusion) attempts to capture this, but no single metric captures it fully.

HLA LOH. Loss of heterozygosity at the HLA locus means the tumor has deleted one of its own MHC alleles. This reduces the range of neoantigens that can be presented and is an active immune escape mechanism. TMB does not account for this.

Practical takeaway

When you see a TMB value in a report:

Note which platform generated it — panel size affects the number
Compare it to the distribution for that tumor type, not to 10 mut/Mb in isolation
Check whether MSI/MMR status was reported separately — it often is, and they are complementary
Remember that TMB predicts at the population level; it is one input into a clinical decision, not a binary answer

The next post covers MHC binding prediction — what NetMHCpan is actually doing when it assigns a percentile rank to a peptide, and how to interpret its output without overreading it.

From the SpatiaBio sister blog

Working with spatial transcriptomics data?

SpatiaBio covers Squidpy-based spatial analysis with runnable notebooks — neighborhood enrichment, co-occurrence, ligand-receptor interactions, and the complete pipeline.

Visit SpatiaBio →

Tags: tumor mutational burden, TMB, pembrolizumab, checkpoint inhibitor, neoantigen, biomarker, KEYNOTE-158, MSI, HLA LOH

NeoantigenLab