Implementing TMB measurement in clinical practice: considerations on assay requirements

Abstract

Clinical evidence demonstrates that treatment with immune checkpoint inhibitor immunotherapy agents can have considerable benefit across multiple tumours. However, there is a need for the development of predictive biomarkers that identify patients who are most likely to respond to immunotherapy. Comprehensive characterisation of tumours using genomic, transcriptomic, and proteomic approaches continues to lead the way in advancing precision medicine. Genetic correlates of response to therapy have been known for some time, but recent clinical evidence has strengthened the significance of high tumour mutational burden (TMB) as a biomarker of response and hence a rational target for immunotherapy. Concordantly, immune checkpoint inhibitors have changed clinical practice for lung cancer and melanoma, which are tumour types with some of the highest mutational burdens. TMB is an implementable approach for molecular biology and/or pathology laboratories that provides a quantitative measure of the total number of mutations in tumour tissue of patients and can be assessed by whole genome, whole exome, or large targeted gene panel sequencing of biopsied material. Currently, TMB assessment is not standardised across research and clinical studies. As a biomarker that affects treatment decisions, it is essential to unify TMB assessment approaches to allow for reliable, comparable results across studies. When implementing TMB measurement assays, it is important to consider factors that may impact the method workflow, the results of the assay, and the interpretation of the data. Such factors include biopsy sample type, sample quality and quantity, genome coverage, sequencing platform, bioinformatic pipeline, and the definitions of the final threshold that determines high TMB. This review outlines the factors for adoption of TMB measurement into clinical practice, providing an understanding of TMB assay considerations throughout the sample journey, and suggests principles to effectively implement TMB assays in a clinical setting to aid and optimise treatment decisions.

Background

Immunotherapy agents have demonstrated significant clinical benefit across multiple tumours; however, as with most therapeutic agents, there is a need for robust biomarkers that predict which patients are most likely to respond to immunotherapy.1 2 Programmed death ligand 1 (PD-L1) expression emerged as a biomarker for immunotherapy,3–5 although challenges with PD-L1 testing exist. PD-L1 immunohistochemistry assays can use different reagents and platforms, and standardisation between them is required.5 There is also uncertainty about whether to assess PD-L1 expression on tumour cells and/or immune cells.6 Furthermore, PD-L1 expression alone is often insufficient for predicting response.7 Despite the challenges, PD-L1 immunohistochemistry assays are approved for companion and complementary testing with immunotherapy.5

Clinical studies to dissect the genetic makeup of tumours revealed that patients with high TMB have increased response rates and improved outcomes to treatment with immunotherapy compared with patients with lower TMB (table 1). This concept was recently validated in a phase III trial with a coprimary endpoint of progression-free survival in patients with high TMB (≥10 mutations/megabase (mut/Mb) measured by the FoundationOne CDx assay8 9).

Table 1.

Summary of clinical evidence demonstrating TMB as a biomarker for response to immunotherapy

Immunotherapy agent and tumour type	Study/trial*	TMB assay used	Type of benefit
Nivolumab
NSCLC (1 L)	CheckMate 02652	WES	ORR, PFS
NSCLC	Flatiron Health117	Foundation CGP panel	OS
Melanoma (1 L or 2 L)	CheckMate 03822	WES	ORR, OS, PFS
Melanoma	CheckMate 06423	WES	ORR, OS
Bladder	CheckMate 27583	WES	ORR, OS, PFS
GBM	Bouffet et al, 2016118	WES	DRR
Ipilimumab
Melanoma	Van Allen et al, 2015119	WES	CBR
	Snyder et al, 201486	WES	CBR, OS
Nivolumab and ipilimumab in combination
NSCLC (1 L)	CheckMate 01235	WES	ORR, DCB, PFS
NSCLC (1 L)	CheckMate 227†8	FoundationOne CDx	ORR, PFS
NSCLC (1 L)	CheckMate 5689	FoundationOne CDx	ORR
SCLC (2 L)	CheckMate 03284	WES	ORR, OS, PFS
Pembrolizumab
NSCLC (1 L)	KEYNOTE-00136	WES	ORR, DCB, PFS
CRC	Le et al, 201544	WES	ORR, PFS
Multiple solid tumours	KEYNOTE-012/KEYNOTE-028120 121	WES	ORR
Atezolizumab
NSCLC (2 L)	POPLAR/OAK87 88	Foundation bTMB	OS, PFS
NSCLC (2 L)	POPLAR/FIR/BIRCH85	FoundationOne	ORR, OS, PFS
NSCLC (1 L)	BFAST and B-F1RST122–124	Foundation bTMB	DOR, ORR, PFS, OS
NSCLC	Rizvi et al, 201856	WES	DCB, ORR, PFS
Bladder (1 L or 2 L)	IMvigor 210125 126	Foundation CGP panel	ORR, OS
		FoundationOne	ORR
Bladder (2 L)	IMvigor 211127	FoundationOne	OS
Bladder	Snyder et al, 2017128	WES	PFS
Multiple agents
NSCLC	Rozenblum et al, 2017129	FoundationOne and Guardant360	ORR
Melanoma	Johnson et al, 201653	FoundationOne	ORR, OS, PFS
	Hugo et al, 201645	WES	OS
Multiple solid tumours	Goodman et al, 2017130	FoundationOne	ORR, OS, PFS
	Yarchoan et al, 201725	Various (not reported)	ORR
Multiple solid tumours (2 L)	Bonta et al, 2017131	FoundationOne	ORR

*Ongoing atezolizumab, durvalumab and avelumab trials have primary completion dates in 2019 and 2020.

†CheckMate 227 has monotherapy and combination therapy arms in the study design.

CBR, clinical benefit rate; CGP, comprehensive genomic profiling; CRC, colorectal cancer; DCB, durable clinical benefit; DOR, duration of response; DRR, durable response rate; GBM, glioblastoma multiforme; NSCLC, non-small cell lung cancer; ORR, objective response rate; OS, overall survival; PFS, progression-free survival; SCLC, small cell lung cancer; TMB, tumour mutational burden; WES, whole exome sequencing.

TMB is defined as the total number of somatic mutations in a defined region of a tumour genome, but the precise definition varies with the sequenced region size and localisation, and the nature of the mutations included.10 11 Testing for genomic alterations, including TMB, is currently performed using next-generation sequencing (NGS). Several NGS approaches exist and the target region ranges from genome-wide analysis (whole genome sequencing (WGS)) to whole exome sequencing (WES, covering the entire coding regions of genes in the genome) and large targeted gene panels (table 1). Many initial clinical studies used WES for TMB assessment, and WES of tumour versus germline DNA is currently considered a reference standard. However, there are many commercial gene panel assays and laboratory-developed tests (LDTs) being used. The US Food and Drug Administration (FDA) recently approved the gene panel assay FoundationOne CDx12 and authorised MSK-IMPACT13 for profiling solid tumours for genomic alterations in a clinical setting, and many other assays are currently available for research use only. As well as testing for actionable mutations, these assays can test for TMB in parallel.14 15 However, they can differ in their methodology (eg, enrichment approach, sequencing system, bioinformatic pipeline), the number of genes and types of mutations included (eg, synonymous and/or nonsynonymous), and the recommendations for interpretation of results. For example, TMB is defined in FoundationOne CDx as the number of somatic, coding, base substitutions (synonymous and nonsynonymous) and short insertions and deletions (indels) per megabase of tumour genome examined.11 14

With a variety of alternative approaches now available, standardisation is required as TMB assays move into clinical practice. It is essential to raise awareness of all assay considerations to ensure that reliable and accurate TMB assessments can be achieved and compared across studies.

Rationale for TMB as a biomarker for response to immunotherapy

Evidence suggests that tumours with higher TMB also carry higher neoantigen loads.16 Tumour-specific somatic mutations can give rise to neoantigens, which are novel protein epitopes specific to tumours that may be presented on the tumour cell surface by major histocompatibility complex molecules.10 17 18 A subset of neoantigens can be recognised as ‘non-self’, leading to T-cell activation and a tumour-targeted immune response.19 20 The clonal nature of tumour evolution implies that T-cell-activating neoantigens will be propagated.21 T-cell receptor sequencing shows that increased clonal neoantigen load may drive clonal T-cell expansion and a sustained antitumor immune response.4 10 22 23

Immune checkpoint molecules negatively regulate T-cell activation, leading to suppression of neoantigen-driven immune responses and allowing tumours to escape immune surveillance.16 Immune checkpoint inhibitors, such as anti-PD-1, anti-PD-L1, and anti-cytotoxic T-lymphocyte antigen 4, restore the antitumour immune response,4 24 and neoantigen-driven activation of the antitumour immune response implies that tumours with high TMB are more likely to respond to immune checkpoint blockade.16 24–26 Moreover, clonal tumour evolution and the effects of increased T-cell activity on other tumour properties may further increase sensitivity to immunotherapy.10 21 27 28

TMB and neoantigen load vary considerably within and across tumour types, with melanomas, lung cancers, and bladder cancers being among those with the highest TMB compared with other tumour types (figure 1).11 29–33 The mutational landscape can also vary across the tumour and evolve during tumour development or following treatment.22 34

Figure 1.

Distribution of TMB and neoantigen load across tumour types. (A) TMB and corresponding predicted neoantigen variation across 14 different tumour types. Data derived from Chen et al.26 (B) TMB variation across 30 different tumour classes. Adapted with permission from Springer: Alexandrov LB, Nik-Zainal S, Wedge DC, et al. Signatures of mutational processes in human cancer. Nature 2013;500(7463):415–21;30 Copyright 2013. AD, adenocarcinoma; ALL, acute lymphocytic leukaemia; AML, acute myeloid leukaemia; CLL, chronic lymphocytic leukaemia; SCLC, small cell lung cancer; SQ, squamous cell carcinoma; TMB, tumour mutational burden.

Tumours with high TMB may exhibit specific genetic alterations associated with clinical benefit to immunotherapy.35 36 For example, high TMB can be caused by defects in mismatch repair (MMR) genes that are associated with microsatellite instability (MSI).11 16 26 The MMR pathway is responsible for the correction of DNA replication errors; such errors frequently occur in microsatellites (short tandem DNA repeats found across the genome). Mutations in MMR genes can lead to increased mutational incidence within microsatellites; MSI can therefore be a surrogate marker for DNA repair disorders, such as deficient mismatch repair (dMMR).37 Similarly, DNA polymerase epsilon (POLE) is required for high-fidelity DNA replication, ensured by Watson–Crick base pairing and exonuclease (proofreading) activity. Mutations in the exonuclease domain of POLE can cause a hypermutator or ultramutator phenotype with TMB >10 mut/Mb or >100 mut/Mb, respectively.38

While clonal tumour expansion may not increase TMB, POLE mutant, dMMR, and MSI-high tumours have been associated with high TMB. However, the reverse is not always true. For example, melanoma and lung tumours frequently have high TMB but <1% are MSI-high39 40 and POLE deficiency occurs in <3% of lung cancers,41 indicating that other mechanisms, including pathogenic events such as UV exposure or smoking, contribute to increased TMB in these tumour types.11 15 38 42 Testing for driver mutations should therefore occur alongside TMB.35 43 Nevertheless, immunotherapy has demonstrated clinical benefit for patients with dMMR and MSI-high tumours,30 44–48 further supporting that TMB, as a surrogate for genome instability as well as tumour neoantigens, is an appropriate biomarker for response to immunotherapy.

Overview of TMB assays and assay variations

NGS approaches

TMB assessment by NGS can involve WGS, WES, or large targeted gene panels, yet analytical and bioinformatic methods are not standardised across research or clinical studies. Both coding and non-coding sequences are analysed in WGS, but because only coding sequences are relevant for TMB assessment, WES, covering the entire exome, and gene panels, covering selected regions, are each suitable.10 16 Gene panels differ in input sample requirement, gene number or identity, region covered, workflow, and bioinformatic algorithms used (table 2); variation in TMB can arise from one or more of these parameters.11

Table 2.

Examples of NGS gene panels in development or currently available to assess TMB

Status	Test name	Number of genes	Coverage (Mb)*	Gene variants	Sample type
FDA-approved or authorised diagnostic assays†	MSK-IMPACT15 56 68	468	1.5	SNVs, indels, rearrangements/fusions, CNAs, parallel analysis of genomic signatures (eg, TMB and dMMR/MSI)	FFPE
	Foundation Medicine FoundationOne CDx14 49	324	0.8	SNVs, indels, CNAs, select rearrangements, parallel analysis of genomic signatures (eg, TMB and dMMR/MSI)	FFPE
Commercial assays for research use only	Caris Molecular Intelligence132	592	1.4	Somatic missense mutations	FFPE
	Illumina TruSight 500 gene panel133	500	2.0	SNVs and indels	FFPE
	Thermo Fisher Scientific Oncomine Tumor Mutation Load Assay77	409	1.7	SNVs	FFPE
	NEO New Oncology NEOplus v2 RUO134	>340	1.1	SNVs, indels, fusions, CNAs, parallel analysis of TMB, MSI, and driver mutations	FFPE
	Foundation Medicine FoundationOne50	315	1.1	SNVs, indels, CNAs, select gene rearrangements, genomic signatures for MSI and TMB	FFPE
	Foundation Medicine bTMB assay88 122	394	1.1	SNVs	Blood
	TruSight Tumor 170135	170	0.5	Fusions, splice variants, SNVs, indels, amplifications	FFPE
	QIAGEN GeneRead DNAseq Comprehensive Cancer Panel97	160	0.7	SNVs, CNAs, indels, and fusions	FFPE
	NEO New Oncology NEOplus105 136	94		SNVs, indels, CNAs, rearrangements, and fusions	FFPE

*Exonic breadth of coverage for the above assays is incomplete because public information may not be available for some assays.

†FoundationOne CDx has FDA premarket approval for mutations associated with several targeted therapies. In addition, FoundationOne CDx can provide tumour mutation profiling to be used by qualified healthcare professionals in accordance with professional guidelines in oncology for patients with solid malignant neoplasms.137 MSK-IMPACT is FDA-authorised to provide information on somatic mutations and MSI. TMB is captured as part of the enhanced report and is considered for investigational use only.13 15

CNA, copy number alteration; dMMR, mismatch repair deficiency; FFPE, formalin-fixed, paraffin-embedded; MSI, microsatellite instability; Mb, megabases; NGS, next-generation sequencing; SNV, single nucleotide variant; TMB, tumour mutational burden.

For most studies to date demonstrating an association between TMB and clinical benefit, TMB was measured by WES. However, it is challenging to routinely implement WES in a clinical setting because this method typically requires complex analysis and a matched normal sample for germline comparisons.11 47 49 50 Interoperability of data by WES is influenced by tumour heterogeneity, artefacts related to tissue preparation of formalin-fixed, paraffin-embedded (FFPE) samples, and discrepancies across assays available from commercial vendors.51

Targeted gene panels have been developed as an alternative method to WES. Whereas WES provides sequence coverage across the entire exome, gene panels are focused on large numbers of cancer-related genes and coupled to bioinformatic algorithms that rapidly report on cancer-targeted genomic alterations, such as copy number alterations (eg, human epidermal growth factor receptor 2 (ERBB2)), gene rearrangements (eg, anaplastic lymphoma kinase (ALK)), MSI, and TMB, along with identification of specific cancer-associated mutations (eg, epidermal growth factor receptor (EGFR^T790M)).13 14 Therefore, comprehensive genomic profiling using gene panels can provide targeted information that may not be immediately available in WES data. Three gene panel tests (FoundationOne, FoundationOne CDx, and MSK-IMPACT) have documented good concordance with TMB assessment by WES using both empirical and in silico approaches.15 52–55 Therefore, gene panels potentially provide an effective approach for TMB estimation in clinical practice.11 15 49 50 56

Further concordance studies are needed to bridge TMB data from WES to gene panels, and translation studies aim to create a framework to define common TMB assay parameters and harmonise data between different gene panels. Ongoing studies by the Quality Assurance Initiative Pathology (QuIP) operated by the German Society of Pathology and the Federal Association of German Pathologists, The Friends of Cancer Research, and the International Quality Network for Pathology (IQN Path) together with the European Society of Medical Oncology (ESMO), are essential for the standardisation of TMB assessment.57–60

TMB panel size

The size of the genome area sequenced differs across assays. WES covers the coding regions of all ~22 000 genes (~180 000 exons, equivalent to 30 Mb), making up ~1% of the genome61 and encompassing most of the known disease-causing variants. FoundationOne CDx and MSK-IMPACT assays cover ~0.8 Mb over 324 genes and ~1.5 Mb over 468 genes, respectively.14 62 Some evidence suggests that smaller gene panels may be sufficient for TMB assessment,63 but other evidence indicates that they may only be sufficient to detect hypermutated and ultramutated tumours and that larger gene panels may be more accurate overall.54 64 Tumours with high TMB may be effectively identified by targeted sequencing of several hundred genes, and accuracy may not increase significantly above this threshold.11 36 56 However, variance increases significantly below 0.5 Mb, particularly in samples with low TMB.11 64 For clinical purposes, evidence suggests that gene panels of ≥0.8 Mb are needed.14 64

It is important to consider whether the specific gene sequences in a panel have an impact on the TMB result. Panels targeted to specific cancer-related genes may introduce unwanted bias into the area of genome covered, but if the panel is sufficiently large, then the function of specific genes within it should be negligible and not impact TMB assessment. There are, of course, driver mutations, such as BRCA1, TP53, and POLE, that are associated with large increases in TMB11 15 38 65 and these should be assessed alongside TMB.46 66

Depth of coverage

Alongside sequencing breadth, sequencing depth (the average number of reads that align to a known reference base67) is also variable across NGS assays. The typical sequencing depth for WES is ~100×, and only mutations with allele frequencies of >15% can be detected with confidence.51 The typical sequencing depth for a gene panel is greater at 500×, increasing the likelihood of detecting low-frequency variants at specific loci,68 69 although the minimum depth of coverage required for accurate TMB assessment may be closer to 200×.70 Together, breadth and depth of coverage may impact the sensitivity and specificity of TMB assessment, but strong evidence suggests that gene panels may be adequate to provide a reliable TMB measurement if sequencing quality thresholds are upheld.

Variant calling

A number of calculations and filters are employed downstream of sequencing to include or discard variants in the TMB estimation. The bioinformatic algorithms used are often not publicly reported, but may differ across assays and may impact TMB output.11 Understanding and harmonising these bioinformatic elements are essential.

One important consideration is the types of mutations included in the assessment, such as insertion and deletion alterations (indels), and base substitutions/single nucleotide variants (which can include synonymous and nonsynonymous mutations).11 15 49 50 71 TMB assessment by WES or gene panels commonly includes nonsynonymous somatic tumour mutations, but these may be restricted to missense mutations only. Recent evidence suggests that outcomes following immunotherapy are more strongly associated with TMB that is defined as including nonsynonymous mutations than with TMB that includes both synonymous and nonsynonymous mutations.35 However, some studies using smaller gene panels may also include synonymous mutations to ensure a representative TMB estimate.10 11 15 30 Another consideration is the variant allele frequency (VAF) cut-offs used to ensure that TMB is a reliable estimation of the number of mutations in the tumour, with minimal artefacts from sequencing errors, formalin-induced DNA damage, or sample contamination.72 73 Variant calling may be improved by capturing and sequencing both DNA strands in the target sequence.74 In studies using WES, VAF can vary from 5% to 10%,75 76 FoundationOne CDx and Oncomine assays call all variants with a VAF ≥5%,14 77 and MSK-IMPACT calls those ≥2% for mutation hotspots and ≥5% for nonhotspots.13 TMB reporting adds a further level of complexity when comparing TMB across studies, as most studies report TMB as mut/Mb, whereas others report TMB as total mutations per tumour.

Somatic variants

It is important to distinguish between tumour-specific acquired (somatic) mutations and those that are present in normal tissue (also known as germline mutations). For WES and MSK-IMPACT, paired sample mutation calling is typically performed on tumour samples and matched germline control (blood) samples.11 13 15 Using control samples for assessment of germline variants may not be feasible, practical, or even legal, because germline testing may have further-reaching implications outside cancer diagnosis.78–80 Somatic mutations can be inferred in tumours without matched controls in silico from databases of known polymorphisms or variant data in tissue from different patients.11 14 54 81 Some assays filter potential germline mutations using a somatic-germline zygosity algorithm, which identifies the origin of the mutation by leveraging allele frequencies and genome-wide copy number.11 14 82 Although these approaches simplify the TMB assessment workflow, they may result in false positives and higher TMB, particularly in patients with ethnic backgrounds that are less well represented in reference databases.11 54 These effects may be negligible for TMB assessment by WES or large gene panels, but increased in smaller panels.54

Defining TMB thresholds

TMB thresholds were initially defined retrospectively in exploratory analyses of immunotherapy efficacy, with patients being grouped by TMB tertile or quartile35 52 83–85 or by numeric cut-offs, such as 100 or 178 mutations per exome22 36 86 or 9–20 mut/Mb.11 87 88 To date, the TMB threshold of ≥10 mut/Mb, measured by FoundationOne CDx (equivalent to ~200 mutations by WES) in first-line patients with non-small cell lung cancer (NSCLC), is the only cut-off established for enhanced response to immunotherapy in one trial and clinically validated in a separate study using a preplanned approach.8 9 55

Identifying a single, fixed TMB threshold that could be applicable across different tumours may be difficult, because the median number of somatic mutations differs across tumour types (figure 1).26 30 In silico analyses suggested that clinical response in NSCLC and melanoma may increase and then plateau at ~260 nonsynonymous mutations (measured by WES),31 but this prediction would need stringent validation to determine which types of mutations contributed to this number and how to translate this value into other patient groups. Further studies evaluating a clinically validated TMB cut-off will be instrumental in providing an agreed TMB cut-off within particular tumour types. Increased use of specific approved commercial assays will also assist in cross-trial comparisons.

Considerations for sample preparation and TMB assays

Samples used for TMB assessment may be cytology or liquid biopsies, or FFPE tumour tissue. Although some evidence exists for the feasibility of using cytology and fine-needle aspiration for analysis by NGS,89 90 these samples can yield variable amounts of DNA that may not always be sufficient for in-depth molecular analysis.91 92 FFPE samples are the most commonly available patient samples for clinical testing. FFPE slides and extracted DNA are reliable for TMB assessment for 6 months and likely beyond.14 However, FFPE samples present a risk for use with high-throughput genomic assays, because preanalytical factors common to NGS testing can affect the sample integrity or amount of DNA that can be extracted.72 73 93 94 These factors include clinical sample availability; sample collection, preparation, and fixation (optimally 24 hours in neutral buffered formalin for surgical specimens and 12 hours for biopsies95 96); neoplastic cell content in the sample; batch processing; and ensuring sufficient DNA quality and yield for assessment (10–100 ng for gene panels and around 250 ng for WES).14 62 97–100 Insufficient DNA is a common cause of sample attrition in clinical trials,52 101 so assays that require less DNA may be advantageous.

In addition to variability in sample handling, technical processes and working practices employed by clinical laboratories can vary extensively.102 Specifically, centralised and decentralised laboratories differ in infrastructure, equipment, and gene panels used, and different platforms may vary in bioinformatic workflow, data integration, and/or how the data are reported.47 103 104

TMB from liquid biopsies

Interest in isolating cell-free DNA (cfDNA) from liquid biopsies for assessment of circulating tumour DNA (ctDNA) is increasing because of the less invasive nature of sample collection from patients and greater opportunity for obtaining frequent samples. TMB measurements from cfDNA may have advantages over tissue biopsy in the follow-up of early responses to immunotherapy due to the ease and speed of obtaining them.105 TMB assays that use cfDNA have been developed and validated, with ongoing efforts for harmonisation with TMB data from tissue samples.87 106–109 Data for TMB concordance between tissue and cfDNA by gene panel assays are conflicting, with reports of both good concordance110 and a lack of concordance.111 Some studies have demonstrated that the sensitivity of detecting mutations from cfDNA is decreased compared with solid tumour samples for both WES and gene panels, although the correlation between TMB assessed from tissue and cfDNA is greater by WES than a panel.75 105 107 Liquid biopsies may yield a variable mix of normal and tumour DNA, and the tumour DNA could reflect heterogeneity across different tumour sites. Thus, variable data exist for concordance between solid and liquid samples, and there is a need for further studies. Nevertheless, future clinical use of TMB assessment from liquid biopsies remains feasible.

Turnaround time

Alongside technical specifications of the tests, other regulatory, practical, and logistic factors are important, such as assay turnaround time, runtime, and cost. Turnaround time is impacted by the time required for steps in the workflow and, therefore, the availability of key stakeholders to carry out their role in the generation of a reliable TMB result (figure 2). Given the comprehensive nature of the assay, it is difficult to balance turnaround time with cost, and microcosting analyses are urgently needed.112 The common turnaround time for TMB assessment by commercial assays is approximately 2 weeks, including 4–5 days of library preparation time.14 69 Moving forward, there is a need to establish acceptable TMB assay turnaround times, especially if the test is applied in the first-line cancer setting. A test for both actionable mutations and TMB is preferable for first-line patients. Moreover, TMB is a surrogate marker for neoantigen load, which may impact immune response and response to therapy, and studies monitoring TMB suggest that follow-up testing could detect residual disease or recurrence.21 22 113

Figure 2.

Biopsy sample workflow for TMB testing. A proposed, optimised workflow is shown to streamline diagnostic testing for TMB alongside other genomic markers. ALK, anaplastic lymphoma kinase; CNA, copy number alteration; EGFR, epidermal growth factor receptor; MSI, microsatellite instability; NGS, next-generation sequencing; QC, quality control; ROS, ROS proto-oncogene 1, receptor tyrosine kinase; TMB, tumour mutational burden.

Quality assurance

Multidisciplinary teams are involved in oncology sampling and diagnostics, from sample collection by the surgeon through to TMB assay result interpretation (figure 2). There is a need for all to be coordinated and well informed on the necessary approaches to achieve a reliable assay result, and molecular tumour boards have been set up to ensure global harmonisation in tumour-sequencing practices.114 Recent NGS guidelines from pathologists emphasise the role of the laboratory director in using an error-based approach that can identify potential sources of errors that may occur throughout the analytical process.72

The robustness of a TMB assessment method, reflected by its approval status (FDA, CLIA, CE-IVD approved, or for research use only), is likely to influence the choice of assay used in clinical practice. Many steps within the TMB assessment pathway can vary and their impact should be assessed. For example, changing an extraction protocol may pose a minor risk if quality control is in place to verify sample purity and concentration; however, changes to which NGS platform is used are likely to require a new validation process.72 Appropriately trained personnel at each step of the workflow to carry out quality-control processes will minimise variance and impact.72

As TMB becomes more established, an increase in the number of commercial or LDT panels is anticipated. Recommendations for how to implement TMB assays will be developed further with input from the international harmonisation efforts described above. It is therefore important to understand how LDTs can augment TMB testing. The identification of reference standards or external quality evaluation is a way to compare LDTs with other methods. It is likely that the clinical evidence for robust TMB assessment will be limited to a few key assays, with a need for calibration and concordance studies to facilitate the transition from clinical trials to the real-world setting.

Challenges and solutions to adopting TMB assays in clinical practice

Because data have demonstrated that high TMB is a predictive marker for clinical benefit from immunotherapy, the ability to classify TMB as high or low is essential for predicting response. Defining cut-offs for TMB may directly lead to inclusion or exclusion of some patients from treatment. However, most published clinical studies of TMB to date have been exploratory, using a number of different methods, and it is challenging to compare data across studies. The availability of preanalytic, sequencing, and bioinformatic protocols will help to harmonise TMB estimation with clinical outcomes.

Lessons can also be learnt from experiences with PD-L1 expression testing. The need for sample quality control, concordance studies across PD-L1 testing platforms to determine the interchangeability of tests, and appropriate training for key personnel to interpret the data are all applicable for any biomarker. Even though TMB outputs are based on computational assessment, there is still a need for education and training on how these values are derived.

With PD-L1 expression and TMB as independent and potentially complementary predictive markers for response to immunotherapy,52 questions turn to the possible need for coordinated multiparameter or composite testing to continue to advance the field. Interplay between TMB, PD-L1, tumour-infiltrating lymphocytes, and cytolytic activity is emerging from comprehensive immunogenomic analyses,26 raising the possibility of composite biomarker testing within the tumour microenvironment.

Conclusions

Validated predictive biomarkers should accurately predict patient responses to immunotherapy. The use of TMB assessment has been actively addressed by ongoing studies, including those clinically validating TMB as a biomarker for response to immunotherapy. Using valuable lessons learnt from PD-L1 testing, including the need for concordance studies across testing platforms and the need for trained personnel to assess the assay results, the development of TMB as a clinical diagnostic tool is being defined.

As the field evolves, solutions emerge to ensure that reliable TMB assessments are carried out in clinical practice. However, definitive recommendations may not be possible until more data are available from prospective clinical trials. From the knowledge gained so far, key parameters can be highlighted throughout the TMB testing workflow, including collection of samples, methods, and analysis (table 3). These include current suggestions to use solid tissue samples, to use gene panels with a coverage >0.8 Mb (preferably >1 Mb), to ensure sufficient information is given when reporting TMB values (including number of genes, capture region, germline subtraction, types of mutations called, variant calling threshold, and sequence depth72 103), to participate in external quality-control schemes when available and to ensure trained stakeholders are available for efficient workflow and assay interpretation.

Table 3.

Key parameters for the harmonisation of TMB analysis and workflow

Parameter	Principles
Sample	Sample must provide adequate quantity and quality of DNA Most cancer NGS assays are performed on FFPE tissue (optimal fixation time, 24 hours for surgical specimens and 12 hours for biopsies). Cytology/FNA, plasma, and other types of samples may be acceptable for smaller targeted panels and are in development to be established for use in the future
Methodology	WES is considered the gold standard for measuring TMB because it offers high breadth of coverage compared with gene panels Investigations of smaller gene panels are ongoing for TMB assessment In general, larger panels have been shown to be more accurate than smaller panels. As more data become available, there may be a recommended number of genes to profile for optimised gene profiling/TMB; current recommendations indicate that the area of genome covered should be >0.8 Mb (preferably >1 Mb) Screening for driver mutations (eg, EGFR, BRAF, KRAS) and TMB in one assay is recommended to avoid time delays or processing of multiple samples. Inclusion of driver mutations, rearrangements, etc on the panel can provide additional clinical utility for the assay beyond TMB, but may not necessarily be informative for the TMB calculation Sequencing depth can affect detection of low-frequency variants. ≥200× is recommended
Platforms	Stakeholders should be aware of varying parameters such as runtime and throughput capacity of available sequencers
Germline	Experimental approaches by filtering germline variants with matched normal samples In silico approaches by filtering against germline variant databases
Algorithm	Standardised and robust bioinformatic pipelines should be developed There is a need for internationally available reference standards (ie, sets of tumours with WES data) Variant allele frequency cut-offs should be reported. Cut-offs of ≥5% are recommended
Cut-off	Currently no standard cut-off to designate high TMB and there is a need for harmonisation of tests to establish set thresholds. More standardised cut-offs are anticipated as clinical utility is established for individual tumour types and TMB thresholds are studied in the context of response to immunotherapies TMB thresholds of ≥13 mut/Mb and ≥10 mut/Mb, measured by FoundationOne CDx, have recently been associated with enhanced responses to immune checkpoint inhibitor monotherapy and combination therapy, respectively, in patients with non-small cell lung cancer. The value cut-off ≥10 mut/Mb was established in one study and clinically validated in a separate trial.8 9 In patients with small cell lung cancer, improved responses were seen with a cut-off of ≥248 missense mutations, as measured by WES84
Concordance	Recommendations on validation processes and concordance testing with WES for approved gene panels or other commercial panels without clinical data Need for direct comparison between panels (especially for FDA-approved or authorised panels)
Reporting	Reporting standards for how TMB is defined and calculated, including gene panel number, capture region, control (germline subtraction), types of mutations called, variant calling threshold, and sequence depth Recommend reporting results as mutations per megabase of targeted DNA Recommend accurate reporting of TMB thresholds and highlight associated clinical outcomes Quality assurance on data reporting is also required

BRAF, B-Raf proto-oncogene, serine/threonine kinase; EGFR, epidermal growth factor receptor; FFPE, formalin-fixed, paraffin-embedded; FNA, fine-needle aspiration; KRAS, KRAS proto-oncogene, GTPase; mut/Mb, mutations per megabase; NGS, next-generation sequencing; TMB, tumour mutational burden; WES, whole exome sequencing.

It will be necessary to accumulate larger datasets, set minimal tissue sample purity thresholds, and standardise sample processing and robust bioinformatic pipelines such that TMB can be calculated in a reliable manner.1 11 99 115 116 Further analyses will also be required to understand how biomarkers, such as TMB, PD-L1, and other genetic/immune markers, interact.

Despite method variability, scoring the number of somatic mutations in tumour DNA is relatively simple, and exhaustive characterisation of tumours will increase understanding of their characteristics. As momentum for TMB as a biomarker of immunotherapy response increases and standardised approaches begin to emerge to allow for clinical implementation, it is likely that TMB assessment, with or without additional biomarkers, will be at the forefront of precision medicine in the foreseeable future.