Tumor mutational burden (TMB) and HLA diversity represent two of the most frequently discussed biomarkers in immunotherapy research, yet they operate at fundamentally different biological levels. TMB quantifies the number of somatic mutations in a tumor genome, serving as a surrogate for neoantigen quantity. HLA diversity captures the germline-encoded capacity of an individual’s immune system to present those neoantigens to T cells. Despite both being implicated in immunotherapy response, they are not interchangeable, and their predictive value is strongest when interpreted together.

Understanding what each metric can and cannot explain is essential for avoiding common interpretive pitfalls. TMB-high does not guarantee an effective antitumor response if the patient’s HLA molecules cannot present the resulting neoantigens. Conversely, optimal HLA diversity does not compensate for a tumor with very few mutations. This article compares these two biomarkers across biological, technical, and practical dimensions to support informed study design and data interpretation in immuno-oncology research.

TMB and HLA Diversity in Immunotherapy Research

Both TMB and HLA diversity have been independently associated with clinical outcomes following immune checkpoint inhibitor therapy, but the strength and consistency of these associations vary substantially across cancer types, treatment contexts, and measurement approaches. Understanding the biological distinction between the two is a prerequisite for designing studies that use either — or both — as predictive or exploratory biomarkers.

TMB reflects the quantity of somatic alterations within a tumor. The rationale linking TMB to immunotherapy response is straightforward: more mutations generate more neoantigens, increasing the likelihood of T cell recognition and antitumor activity. In cancers with high mutational loads — melanoma, non-small cell lung cancer, and microsatellite-unstable colorectal cancer — the association between TMB and checkpoint inhibitor response is reasonably robust. In cancers with low mutational burden, including most breast, prostate, and pancreatic cancers, the relationship is weaker and often not statistically significant.

HLA diversity, by contrast, reflects the germline-encoded capacity of an individual’s antigen presentation machinery. Each person inherits up to six classical HLA class I alleles (two copies each of HLA-A, HLA-B, and HLA-C), and the evolutionary distance between these alleles determines the breadth of peptides that can be presented. Greater HLA diversity — measured as heterozygosity at each locus, sequence divergence between alleles, or both — increases the range of neoantigens displayed to T cells. Unlike TMB, which varies between tumor samples over time, HLA genotype is stable and can be determined from any nucleated cell source.

The practical implication for research teams is that TMB and HLA diversity address different questions. TMB asks “how many potential targets exist?” HLA diversity asks “how many of those targets can the immune system see?” Both questions are relevant, but neither alone provides a complete answer.

Mutation Load as a Surrogate for Immunogenicity

TMB is most commonly measured by whole exome sequencing or targeted panel sequencing, with results reported as mutations per megabase (mut/Mb). The FDA-approved threshold of 10 mut/Mb for pembrolizumab in solid tumors has established a binary framework, but the relationship between mutation count and immunogenicity is more nuanced than a simple cutoff implies.

The biological premise of TMB rests on the assumption that more mutations produce more peptide neoantigens capable of binding HLA molecules and eliciting T cell responses. This holds reasonably well at the upper end of the mutational spectrum:

• Hypermutation phenotypes — MSI-H, POLE/POLD1 mutations, or mutagen exposure generate hundreds to thousands of somatic mutations, and a fraction produce immunogenic neoantigens
• Extreme contexts — TMB’s predictive value is strongest in melanoma, NSCLC, and MSI-H colorectal cancer, where mutation counts are highest
• Diminishing returns — across the broader patient population where mutation counts fall near conventional thresholds, predictive value diminishes considerably (Niknafs et al., 2025)

A 2024 review by Wang and colleagues in Annals of Oncology systematically examined challenges limiting TMB as a predictive biomarker. The authors identified several sources of variability: differences between whole exome and targeted panel measurements, the impact of tumor purity on allele frequency calculations, and confounding effects of clonal hematopoiesis when using blood-based TMB assessment. These technical factors contribute to discordance rates between TMB measurements from different platforms that can exceed 20%, even when the same tumor sample is analyzed.

The more fundamental limitation is that mutation quantity does not equal neoantigen quality. A tumor with 15 mut/Mb may produce zero immunogenic neoantigens if all mutations occur in genomically silent regions, are not expressed as protein, or generate peptides that cannot bind the patient’s HLA molecules with sufficient affinity. Researchers designing studies that rely on TMB as a primary biomarker should consider complementing it with HLA-based analyses that assess the presentation step of the neoantigen pathway.

Antigen Presentation Capacity as a Separate Biological Layer

HLA diversity captures a dimension of immune biology that TMB cannot access: the capacity of the antigen presentation system to display actionable targets to the immune system. This capacity is determined by three main factors:

• Number of distinct HLA alleles — Individuals heterozygous at all three class I loci carry six distinct HLA molecules compared to three for homozygotes. Each additional molecule expands the neopeptide repertoire and increases the pool of potential T cell targets
• Sequence divergence between alleles — Greater evolutionary distance between paired HLA alleles increases the range of peptides that can be presented, as measured by divergence-based metrics like HAPS
• Functional status of HLA genes — HLA loss of heterozygosity (LOH) and B2M mutations can eliminate or reduce antigen presentation capacity even when germline diversity is favorable

HLA heterozygosity at each class I locus provides an immediate advantage for neoantigen presentation. At the population level, HLA heterozygosity has been associated with improved survival following checkpoint blockade across multiple cancer types.

A 2024 study by Han and colleagues in Nature Communications developed the HLA tumor-Antigen Presentation Score (HAPS), which integrates neoantigen binding affinity predictions with the evolutionary divergence between paired HLA alleles. The study demonstrated that HAPS outperformed TMB, PD-L1 expression, and HLA heterozygosity alone for predicting immunotherapy benefit across 885 pan-cancer patients. When HAPS was combined with TMB and TCR diversity in a neural network model, the integrated score provided the strongest stratification of overall and progression-free survival. This finding directly illustrates the complementary nature of TMB and HLA information: TMB provides the raw material, HLA diversity determines how much of that material is functionally available.

HLA loss of heterozygosity (LOH) represents a mechanism by which tumors actively subvert antigen presentation. A large-scale 2025 pan-cancer analysis of over 48,000 tumors published in Annals of Oncology found that approximately 15% of tumors carried HLA class I LOH. In tebentafusp-treated melanoma patients, baseline HLA-A*02 LOH was associated with a median progression-free survival of 1.4 months compared to 6.5 months in patients without LOH (HR 6.8), and a corresponding survival difference of 5.5 vs 19.8 months. HLA LOH screening is increasingly incorporated into study designs for HLA-restricted therapies.

Where TMB Falls Short: Heterogeneity and Thresholds

The limitations of TMB as a standalone biomarker have been well-documented and fall into three categories:

• Technical heterogeneity — Differences in panel design, sequencing depth, bioinformatic pipelines, and germline filtering strategies mean a TMB value of 8 mut/Mb from a 500-gene panel is not directly comparable to a value from WES. Bioinformatic choices can shift TMB values by several mut/Mb for the same sample. A 2025 review by Mouawad and colleagues documented that discordance between TMB assays is a persistent barrier to clinical translation, with cancer-type-specific thresholds further complicating interpretation
• Biological heterogeneity — A single biopsy may not capture the mutational landscape of a heterogeneous tumor, and TMB measured from a metastatic biopsy may differ substantially from the primary tumor. Only clonal mutations present in all tumor cells contribute consistently to neoantigen presentation; subclonal mutations restricted to a subset of cells have proportionally smaller effect
• Threshold ambiguity — The optimal TMB cutoff varies by cancer type, treatment regimen, and clinical endpoint. The pan-cancer threshold of 10 mut/Mb does not perform equivalently across indications. In MSS CRC, the association between TMB and immunotherapy benefit is weak or absent even at high mutation counts, with POLE/POLD1 mutations and TME features being more informative

A comparison of key differences between TMB and HLA diversity as biomarkers:

Where HLA Diversity Adds Independent Information

HLA diversity contributes predictive information that is orthogonal to TMB in several important contexts. Even in TMB-high tumors, the ability to present the resulting neoantigens depends on the patient’s HLA genotype, and tumors can actively evade this step through HLA LOH or beta-2-microglobulin (B2M) loss.

A 2024 study by Reis and colleagues in Frontiers in Immunology tracked B2M and HLA-A expression dynamics in tumors during immunotherapy. Up to 56% of tumors showed B2M or HLA-A loss, and on-treatment increases in B2M expression correlated with clinical response while baseline expression levels did not. This finding has direct implications for biomarker study design: a single pre-treatment measurement of TMB or HLA genotype may miss the dynamic regulation of antigen presentation that occurs during therapy. Incorporating longitudinal assessment of HLA pathway components alongside TMB provides a more complete picture of the immune recognition axis.

The 2024 Nature Medicine study by Kinget and colleagues developed a spatial architecture-embedding HLA signature for renal cell carcinoma that integrated HLA gene expression patterns with their spatial organization within the tumor microenvironment. The signature identified a crosstalk between proinflammatory tumor-associated macrophages and exhausted CD8+ T cells that correlated with a neoantigen-favoring HLA repertoire. This spatial dimension — the distribution of HLA-expressing cells relative to T cells — represents an additional layer of information that neither TMB nor simple HLA genotyping captures.

HLA analysis adds value across the full TMB spectrum. In TMB-high tumors, HLA diversity and LOH status determine whether available mutations can be functionally presented. In TMB-low tumors, favorable heterozygosity may partially compensate for lower mutation counts. HLA typing sequencing approaches that provide allele-level resolution and LOH detection enable researchers to incorporate this dimension into their biomarker analyses.

Technical Considerations for TMB and HLA Measurement

The technical requirements for TMB and HLA measurement differ substantially, and study designs must account for the specific needs of each.

For TMB, the choice between whole exome sequencing and targeted panel sequencing affects both the accuracy and precision of the measurement. WES provides a genome-wide assessment of coding mutations and remains the reference standard, but it requires higher sequencing depth and bioinformatic resources. Targeted panels optimized for TMB estimation typically include 300–500 genes, with the number of coding bases determining the confidence interval of the TMB estimate. Cancer panel sequencing approaches designed for TMB assessment incorporate variant filtering against germline databases and clonal hematopoiesis correction to improve specificity.

For HLA typing, NGS-based methods have largely replaced PCR-based typing for research applications. These methods sequence the polymorphic exons of HLA genes (primarily exons 2 and 3 for class I, exon 2 for class II) and achieve allele-level resolution sufficient for neoantigen binding prediction. HLA typing can be performed from the same DNA samples used for TMB measurement — either from tumor tissue with matched normal, or from blood samples — without requiring additional tissue collection. The critical technical requirement is that both TMB and HLA measurements are performed on samples that accurately represent the patient and tumor.

For researchers integrating both biomarkers, a practical approach is to design the sample collection workflow to support both analyses from a single set of specimens. Blood or buffy coat provides the germline DNA needed for HLA typing and for filtering germline variants from TMB analysis. Tumor tissue (fresh, frozen, or FFPE) provides the somatic DNA for TMB measurement. When both are collected in parallel, the incremental technical effort for adding HLA analysis to a TMB-focused study is relatively modest.

Integrating TMB and HLA Data in Research Design

The studies described above converge on a consistent finding: TMB and HLA diversity provide complementary information, and their combined use outperforms either biomarker alone. For research teams designing biomarker studies, several practical approaches to integration are available.

The simplest integration uses TMB as a filter and HLA analysis as a secondary layer. Studies can stratify by TMB status and assess whether HLA diversity, heterozygosity, or HAPS score further differentiates outcomes within strata.

A more integrated approach uses predictive models combining TMB with HLA-derived features. The HAPS score provides a validated framework combining neoantigen binding predictions with HLA allele divergence, further improved by incorporating TMB. For studies with access to multiplex immunofluorescence or spatial transcriptomics data, adding spatial metrics of HLA expression and T cell infiltration provides additional resolution.

For studies investigating immunotherapy mechanisms of action, longitudinal assessment of both TMB (through ctDNA or serial biopsies) and HLA pathway components (through B2M and HLA expression analysis) can reveal how the neoantigen presentation axis evolves during treatment. Tumor microenvironment profiling services that integrate genomic, immunogenetic, and spatial data increasingly enable these multi-dimensional study designs.

The key design principle is that TMB and HLA diversity should not be treated as redundant biomarkers. They measure distinct biological steps in the neoantigen recognition pathway, and their combined assessment provides a more complete picture of the factors determining immunotherapy response. For research teams, the cost of adding HLA analysis to an existing TMB-focused study is modest relative to the additional explanatory power it provides.

FAQ

Can TMB and HLA diversity be measured from the same sample?

Yes. Blood or buffy coat provides the germline DNA needed for HLA typing and for filtering germline variants from TMB analysis. Tumor tissue provides the somatic DNA for TMB measurement. When both sample types are collected in parallel, the incremental technical effort for adding HLA analysis to a TMB-focused study is relatively modest.

Why doesn’t high TMB always predict immunotherapy response?

Mutation quantity does not equal neoantigen quality. A tumor with 15 mut/Mb may produce zero immunogenic neoantigens if all mutations occur in genomically silent regions, are not expressed as protein, or generate peptides that cannot bind the patient’s HLA molecules with sufficient affinity. Additionally, TMB does not account for HLA LOH, B2M loss, or other mechanisms of antigen presentation evasion.

What is HLA loss of heterozygosity and why does it matter?

HLA LOH is a somatic event where a tumor cell loses one copy of an HLA gene, reducing the number of available HLA molecules for neoantigen presentation. It occurs in approximately 15% of tumors across cancer types. In tebentafusp-treated melanoma patients, baseline HLA-A*02 LOH was associated with median PFS of 1.4 months vs 6.5 months without LOH (HR 6.8).

How do I calculate HAPS for my research cohort?

The HLA tumor-Antigen Presentation Score (HAPS) integrates neoantigen binding affinity predictions with the evolutionary divergence between paired HLA alleles. It requires HLA typing data (allele-level resolution), tumor mutation data (for neoantigen prediction), and computational tools for binding affinity prediction. Han and colleagues (2024) provide the validated framework in Nature Communications.

Is HLA typing from NGS data as reliable as dedicated HLA typing assays?

NGS-based HLA typing methods have largely replaced PCR-based typing for research applications. These methods sequence the polymorphic exons of HLA genes (exons 2–3 for class I, exon 2 for class II) and achieve allele-level resolution sufficient for neoantigen binding prediction. The accuracy depends on sequencing depth, bioinformatic pipeline, and whether paired tumor-normal data is available for LOH detection.

Key Takeaways for Biomarker Research

• TMB measures neoantigen quantity (mutation supply), while HLA diversity measures antigen presentation capacity (immune access) — these are distinct biological layers that require different measurement approaches
• TMB’s predictive value is strongest in high-mutation contexts (MSI-H, POLE-mutant, melanoma, NSCLC) and weaker in low-mutation tumors (breast, prostate, pancreatic) and MSS CRC
• HLA heterozygosity, sequence divergence, and functional status (LOH, B2M expression) provide independent predictive information across the full TMB spectrum
• Combined models integrating TMB with HLA features (HAPS, HLA LOH status, spatial HLA signatures) consistently outperform single-biomarker approaches
• Technical considerations differ substantially between TMB and HLA measurement, but both can be incorporated into a single sample collection workflow
• HLA LOH occurs in approximately 15% of tumors and is a strong negative predictor for HLA-restricted therapies, supporting its inclusion in biomarker study designs
• For teams new to HLA analysis, adding allele-level HLA typing to an existing TMB-focused study requires modest effort and provides substantial interpretive value

For Research Use Only. Not for use in diagnostic procedures.