TMB measurement, validation and pre-analytical considerations

In general, the incorporation of new cancer biomarkers, particularly those that need enough tissue, into routine clinical practice is very demanding since it should be backed up with strong clinical evidence. In addition, the test should be performed with a minimal amount of DNA, have a reasonable cost to be reimbursed and turnaround time that do not significantly delay therapeutic interventions, and provide accurate results. This is even more challenging with TMB, owing to its complex NGS workflow and the need for in-depth bioinformatics expertise. TMB estimation needs larger amount of high-quality DNA than those for single gene testing, WGS requires between 50 nanograms and 1 microgram of high-quality DNA and therefore it is critical to obtain enough tissues to overcome this issue and address tumour heterogeneity and avoid false-negative results. It is not only about quantity but also the quality of the DNA is even more important. Moreover, there should be an adequate percentage of viable tumour nuclei within the sample. For a single-gene testing tools such as Sanger sequencing, 40% of tumour DNA is enough for the detection of variants; however, for WGS which includes broader and more comprehensive coverage, so a larger genetic content is required. Therefore, The Cancer Genome Atlas (TCGA) excludes tissues with 20–50% necrosis and necessitates samples with greater, 60–70%, tumour nuclei, this criterion is even stricter for glioblastoma multiforme (GBM) and requires 80% tumour nuclei. One solution is to improve sample quality by dissecting and removing necrotic areas before analysis (The Cancer Genome Atlas Research Network 2008). The current process for DNA fixation is the formalin-fixed, paraffin-embedded (FFPE), which is associated with many drawbacks, has that can lead to DNA damage. Instead, recent studies have considered Fresh Frozen (FF) for tissue fixation and preservation to overcome formalin damage. Although FF has also several issues but primarily logistical related to storage at ultralow temperature, using liquid nitrogen (LN), which is extremely expensive, and such infrastructure is not widely available in hospitals. Moreover, there is risk of sample damage in case of temperature changes and also serious risks, such burns, tank explosions, and suffocation in case of LN2 leakage. Most importantly, FF provides high quality DNA compared to FFPE (FFPE vs Frozen Tissue Samples 2018; Fresh vs Frozen Samples: Human Clinical Samples 2018; Robbe et al. Reference Robbe2018).

Another challenge in tissue sample-based assays is the tumour heterogeneity which refers to the presence of genetic and phenotypic discrepancies within a tumour or between different regions of the same tumour which can impact TMB estimation in several ways. Subclonal mutations: Tumours often contain subpopulations of cells with different genetic profiles, where some mutations may be present in only a small fraction of tumour cells. This can lead to an underestimation of TMB if the assay does not capture all the subclonal mutations. Second, spatial heterogeneity: different regions of a tumour may have distinct mutation profiles; therefore, the biopsy of a single region may not capture the most mutated region, leading to an inaccurate estimation of TMB. Third, temporal heterogeneity: Tumours can evolve over time, acquiring new mutations or losing existing ones; thus, a single biopsy may not capture the full spectrum of mutations present at different stages of tumour development. This can lead to variability in TMB estimation if relying on an archived tissue that does not align with the most recent mutational load (Schmelz et al. Reference Schmelz, Toedling and Huska2021).

In recent years, the analysis of circulating tumour DNA (ctDNA), commonly referred to as liquid biopsy, has undergone substantial advancements. This methodology possesses significant potential to address numerous challenges previously outlined. The sequencing of ctDNA yields critical insights into the dynamics of the oncogenic mutational landscape. Furthermore, it serves as a real-time biomarker that facilitates the accurate and timely assessment of TMB. Additionally, liquid biopsy offers a noninvasive tool for the continuous monitoring of therapeutic responses, evaluation of minimal residual disease, and early detection of disease progression indicators (Sivapalan et al. Reference Sivapalan, Murray and Canzoniero2023).

Variation in breadth and depth of coverage

The genome coverage varies according to the assay or platform. Whole exome sequencing (WES) covers only the protein coding regions, accounting for approximately 1–2% of the human genome, and around 22,000 genes with 30–50 Mb in size. Thus, WES has the capacity to detect most of the genetic variants associated with diseases. In contrast, targeted gene panels cover a smaller range of size and number of genes, for example, FoundationOne CDx covers a total 0.8 Mb and 324 genes, while the MSK-IMPACT assay covers a total of 1.5 Mb and 468 genes. Clinical studies have indicated that gene panels smaller than these may be insufficient for accurate TMB estimation. Inconsistent TMB measurements have been associated with panels covering <0.5 Mb of the genome. Gene panels of ≥0.8 Mb are therefore essential for the accurate TMB estimation (Ng et al. Reference Ng, Turner and Robertson2009; Baras et al. Reference Baras and Stricker2017; Evaluation of Automatic Class III Designation for MSK-IMPACT (Integrated Mutation Profiling of Actionable Cancer Targets): decision summary 2018; FDA 2018b). The depth of sequencing is also important and it too varies significantly based on the various NGS assays or the platforms used. The minimum coverage depth required for precise TMB estimation is around 200×. However, WES provides ~100×, and can only detect mutations with allele frequencies >15%. In contrast, gene panels provide deeper coverage at approximately 500×, which improves the detection of low-frequency variants. Therefore, gene panels can provide adequate coverage and reliable TMB estimation (Cheng et al. Reference Cheng, Mitchell and Zehir2015; Feliubadaló et al. Reference Feliubadaló, Tonda and Gausachs2017; Lee et al. Reference Lee, Bae and Ryu2017).

Variation in TMB estimation

The TMB estimation varies based on multiple factors, including the NGS platforms, panel size, depth of coverage, somatic variants/mutations counted, and TMB threshold. In the meantime, a standardised method for TMB analysis, interpretation, and result reporting remains undetermined. A recent study by the Quality in Pathology (QuIP) reported that up to 25% of samples had been misclassified as TMB-H and TMB-L. The laboratories included in this study utilised various TMB methods, including commercially available techniques such as Oncomine™, while other centres developed their own panels for TMB estimation. Moreover, the type of mutations considered for TMB detection and cutoff TMB values used for result interpretation also varied significantly between the participating laboratories. Collectively, such discrepancies led to inconsistent interpretations of the results, negatively impacted the clinical utility, and limited the widespread utilisation of TMB as a predictive biomarker. Furthermore, 19 laboratories used cell-free DNA (cfDNA) to quantify TMB, despite of the limited evidence on its sensitivity and specificity for TMB testing, as well as the very low allelic frequency of variants that could be detected in the peripheral blood (Fenizia et al. Reference Fenizia, Pasquale, Roma, Bergantino and Iannaccone2018; Gandara et al. Reference Gandara, Paul and Kowanetz2018; Stenzinger et al. Reference Stenzinger, Endris, Budczies, Merkelbach-Bruse and Kazdal2020). These findings raise serious concerns on the reproducibility of TMB results and reinforce the urgent need for standardisation, validation, and clinical accreditation of TMB. Additionally, the Friends of Cancer Research (FoCR) TMB Harmonisation Project study has reported that filtering out the pathogenic variants is critical to avoid the overestimation of TMB. Table 1 summarises the various types of the available TMB assays.

Table 1.

Summary of the various available assays and platforms for TMB estimation

Differences in NGS approaches or platforms

There are different workflows that can be used for TMB analysis: WGS, WES, or large targeted gene panels, and each has its advantages and disadvantages. The WGS Workflow provides the most comprehensive since it covers the entire genome. Thus, it can detect almost all types of genetic variants which lead to the most accurate estimation of TMB. However, it requires the highest sequencing depth and coverage and generates large amounts of data, requiring more computational resources for analysis, subsequently it is the most expensive and resource-intensive workflow. WES Workflow is regarded as the gold standard method for TMB assessment and has been extensively used in clinical trials that demonstrated an association between TMB response and the clinical efficacy of ICI treatment. Since it can provide a more accurate and comprehensive estimation of TMB due to its higher sequencing depth and broader coverage of the exome, capturing a broad range of variants, including SNVs, indels, as well as CNVs. However, its incorporation into routine clinical settings is challenging and rather reserved for research purposes, as it requires complex analysis and matching with a normal DNA sample to eliminate germline variants, thus accounting for the somatic genetic aberrations only, and may lead to potential false-negative results in poorly covered regions. Therefore, it is still associated with long turnaround time, high operational costs, and complex bioinformatics for data analysis and interpretation (Abbasi et al. Reference Abbasi and Alexandrov2021; Pei et al. Reference Pei, Yeung and Wong2023). Targeted Gene Panel Workflow is also considered a potentially acceptable and reliable way for TMB estimation in clinical practice since it focuses only on a specific subset of cancer-related genes that are known to be more relevant to the tumour biology, allowing for deeper sequencing and higher coverage, and therefore, it’s more cost-effective than WES. Thus, large targeted gene panels have been routinely utilised in the clinical settings, and several commercially available targeted gene panels can be used for the TMB quantification. On the contrary, it may potentially miss variants in non-targeted regions, and leading to an underestimation of TMB. Moreover, gene panels vary in terms of the input sample needed, the number of genes and the genes included, the regions covered, the methodology, and the bioinformatics methods. These factors may contribute to discrepancies in the estimation of TMB and, ultimately, its predictive value (Frampton et al. Reference Frampton, Fichtenholtz and Otto2013; Chalmers et al. Reference Chalmers, Connelly, Fabrizio, Gay, Ali and Ennis2017; Allgäuer et al. Reference Allgäuer, Budczies, Christopoulos, Endris, Lier, Rempel, Volckmar, Kirchner, von Winterfeld, Leichsenring, Neumann, Fröhling, Penzel, Thomas, Schirmacher and Stenzinger2018; FDA unveils a streamlined path for the authorization of tumor profiling tests alongside its latest product action 2018; Meléndez et al. Reference Meléndez, Van Campenhout, Rorive, Remmelink, Salmon and D’Haene2018; Büttner et al. Reference Büttner, Longshore, López-Ríos, Merkelbach-Bruse, Normanno, Rouleau and Penault-Llorca2019; Stenzinger et al. Reference Stenzinger, Endris, Budczies, Merkelbach-Bruse and Kazdal2020; Meri-Abad et al. Reference Meri-Abad, Moreno-Manuel and García2023; Zhang et al. Reference Zehir, Benayed, Shah, Syed, Middha and Kim2024). Therefore, concordance studies are required to provide a standardised framework, to harmonise data between various gene panels, and translate TMB data from WES into gene panels.

Somatic mutations and variant calling

Variant calling is also a significant variable in determining the TMB. Various bioinformatics methods or filters are employed to include or exclude certain genetic variants from the TMB assessment. Moreover, there are different types of mutations considered for TMB estimation, such as single nucleotide variants (SNVs) consisting of both synonymous and nonsynonymous mutations, as well as small insertions and deletions (indels). These factors are vital and should be taken into account as they have a direct and significant impact on TMB results (Singh et al. Reference Singh, Patel and Routbort2013; Koeppel et al. Reference Koeppel, Blanchard and Jovelet2017; Hellmann et al. Reference Hellmann, Ciuleanu and Pluzanski2018; Sung et al. Reference Sung, Wang and Li2022). WES and NGS gene panels mainly detect SNVs in tumours, thus limiting estimation of TMB and the neoantigen repertoire to missense and nonsense mutations. Although recent studies have demonstrated that responses to immunotherapy are more closely associated with nonsynonymous than synonymous mutations, TMB estimation often does not distinguish between these types of mutations, only the number of SNVs.

There are several steps involved in the calculation of targeted panel-based TMB; first, variant calling and defining the true variants based on quality metrics then the annotation of variant types included for TMB estimation. Second, the filtration of germline mutations and single-nucleotide polymorphisms (SNPs) to be excluded from TMB calculation. Third, the deployment of an algorithmic adjustment to reduce or eliminate the bias of cancer hotspot mutations. Finally, the use of regression model to validate the TMB estimation methodology (Lauss et al. Reference Lauss, Donia and Harbst2017).

The variant allele frequency (VAF) threshold also varies across NGS panels and TMB platforms. While WES captures variants with VAFs of 5–10%, FoundationOne CDx and Oncomine assays detect variants with a VAF of ≥5% and MSK-IMPACT panel detects hotspot mutations with a VAF of ≥2% and non-hotspot mutations with a VAF of ≥5% (Srinivasan et al. Reference Srinivasan, Sedmak and Jewell2002; Jennings et al. Reference Jennings, Arcila and Corless2017; Riaz et al. Reference Riaz, Havel and Makarov2017; FDA 2018b; ThermoFisher Oncomine™ tumor mutation load assay user guide 2018). Moreover, errors in TMB estimation occur due to formalin fixation of samples. DNA damage, artefacts, or sample contamination may all contribute to the overall TMB estimation. To overcome this issue and enhance variant calling, sequencing of both DNA strands is advised (Rizvi et al. Reference Rizvi, Hellmann and Snyder2015; Snyder et al. Reference Snyder, Makarov and Merghoub2014). Furthermore, TMB estimation becomes complex in terms of its measurement units (mut/Mb versus total mutations/tumour) while comparing the TMB across various studies.

TMB thresholds for diverse tumour types

TMB is a continuous and even dynamic variable. Differences ranging from 0.001/Mb to >1,000/Mb have been observed across various cancers and even within the same cancer type. Cancers developing in response to chronic exposure to carcinogens, such as melanoma, UV light, and lung cancer to tobacco, exhibit some of the highest TMBs. In contrast, TMB has been found to be low in paediatric, gastrointestinal, and haematological malignancies, whereas breast, kidney, and gynecologic cancers exhibit intermediate TMB levels. The TMB variation is observed not only across different tumour types, but also across different histological subtypes within the same cancer type. For example, lung, head, and neck cancers exhibit less variation in TMB, whereas colon, urothelial, and endometrial cancers show greater TMB heterogeneity (Alexandrov et al. Reference Alexandrov, Nik-Zainal, Wedge, Aparicio, Behjati and Biankin2013; Chalmers et al. Reference Chalmers, Connelly and Fabrizio2017; Zehir et al. 2017; Vanderwalde et al. Reference Vanderwalde, Spetzler, Xiao, Gatalica and Marshall2018; Merino et al. Reference Merino, McShane, Fabrizio, Funari, Chen and White2020). The difference in the prevalence landscape of TMB across various cancer types is shown in Figure 3.

Figure 3.

Landscape of TMB prevalence across various cancer types.

The initial TMB quantification was based on a retrospective exploratory analysis of randomised ICB trials, which used numeric cutoffs of either 178 muts/exome (WES assessment) or 10–20 mut/Mb (targeted gene panels) (Mellman et al. Reference Mellman, Coukos and Dranoff2011; Fabrizio et al. Reference Fabrizio, Lieber and Malboeuf2018; Gandara et al. Reference Gandara, Paul and Kowanetz2018; Hellmann et al. Reference Hellmann, Nathanson and Rizvi2018; Ramalingam et al. Reference Ramalingam, Hellmann and Awad2018; Szustakowski et al. Reference Szustakowski, Green and Geese2018). Meanwhile, the most extensively studied and clinically validated approach (prospectively) was used for NSCLC, in the clinical trials of checkmate-568 and checkmate-227, in which the TMB threshold of ≥10 mut/Mb estimated by FoundationOne CDx was established. The determination of a universal TMB threshold that can be used across various cancer types is unlikely, owing to the significant variation in the median number of somatic mutations across tumour types (Blank et al. Reference Blank, Haanen, Ribas and Schumacher2016; Chen et al. Reference Chen, Zhang and Lv2017; Goodman et al. Reference Goodman, Kato and Bazhenova2017; Galanina et al. Reference Galanina, Goodman and Cohen2018). Thus, further research is required to accurately determine the clinically validated TMB thresholds for each cancer type.

Some TMB-L tumours respond to ICIs

Another confounding issue is that, although TMB-H has been correlated with vulnerability to ICI therapy, some patients with TMB-L respond to ICIs (Turajlic et al. Reference Turajlic, Litchfield and Xu2017). For instance, many patients with Kaposi sarcoma achieved complete or partial responses when treated with PD-1 antibodies despite a low TMB (Saeterdal et al. Reference Saeterdal, Bjørheim, Lislerud, Gjertsen, Bukholm, Olsen, Nesland, Eriksen, Møller, Lindblom and Gaudernack2001). This result raises questions regarding the role of TMB as a biomarker for the selection of patients who receive immunotherapy, and several questions remain unanswered confounding these results. First, how confident are we in the false negativity of TMB, the heterogeneity across various NGS panels, and the vast technical requirements to accurately run the TMB tests? Second, patients who respond to ICIs often have tumours with a large number of tumour infiltrating lymphocytes (TILs). Thus, the biopsy specimens available for such samples might contain an insufficient proportion of tumour cells relative to TILs, thereby leading to a false-negative or inaccurate TMB status. Finally, some of these studies relied on archival tissues, which might not be representative of the actual genetic status of these patients at the time of treatment.

Furthermore, gene alterations affecting other molecules in the immune response pathway may obscure the significance of the TMB estimation. For example, a recent study demonstrated that tumours with loss of heterozygosity for HLA (HLA-LOH) exhibit higher TMBs compared with tumours without HLA-LOH. However, the downregulation of HLA genes is an immune evasion strategy for cancer cells. Loss of heterozygosity (LOH) in the HLA-I alleles, a total of six different HLA-I alleles at three loci, HLA-A, HLA-B, and HLA-C, is observed in various cancers and has been associated with poor outcome in response to ICIs. A computational tool was recently developed enabling the quantification of the allele-specific copy number of the HLA locus. These algorithms have been shown to help better classify patients into TMB-H and TMB-L groups, and it was found that the HLA-corrected TMB has better predictive power for PFS and OS (McGranahan et al. Reference McGranahan, Rosenthal and Hiley2017; Shim et al. Reference Shim, Kim and Cha2020). Thus, HLA-corrected TMB can also help to better predict patients with TMB-H who will not respond to ICIs.