Tumor mutational burden: why is it still a controversial agnostic immunotherapy biomarker?

ABSTRACT

For the past few years, researchers and oncologists have been pushing to find biomarkers that would help predict which treatment option would best work on a patient. Tumor Mutational Burden (TMB) is one of the latest biomarkers that is being studied and considered as a promising agnostic immunotherapy biomarker. However, it still shows controversial results in studies due to the difficulty in finding solid comparable results. This is a consequence of different cutoff definitions among many cancer types, age ranges, and the use of different sequencing assays, in addition to its association with other biomarkers such as PD-L1. Finally, the use of composite biomarkers to assess the genetic signature of a tumor might be the way forward to seriously use TMB as an agnostic biomarker.

1. Introduction

Tumor mutational burden (TMB) is known as the total number of somatic mutations per megabase in a definite tumor genome region [1]. It is being widely studied and tested as it is starting to be considered a promising pantumor biomarker and showing positive results regarding its predictive value of response rate to immune checkpoint inhibitor (ICI) therapy [2,3]. The believed mechanism, illustrated in Figure 1, behind this biomarker is that a tumor with a high tumor mutational burden (TMB-H) produces high-quality neo-antigens that increase T-cell reactivity, which would lead to a better response to immune checkpoint inhibitors [4,5]. However, TMB has yet to be a completely understood, well-defined concept to become a useful clinical tool for oncologists as an agnostic biomarker, as opinions about TMB among specialists are still divergent. We conducted this review to help assess if TMB can be used as an agnostic biomarker, and if not, what is still needed to bridge this gap.

Figure 1.

The mechanism of TMB. Tumors with high TMB secrete a high amount of neoantigens, which increases T-cell reactivity thus leading to a better response to immune checkpoint inhibitors.

2. Association between TMB levels and response to immune checkpoint inhibitors

Many studies and clinical trials, illustrated in Tables 1 and 2, explore the association between TMB levels or PD-L1 status and response to treatment. For instance, Hanna et al. showed that TMB-H tumors were associated with improved outcomes in the patients when treated with anti-PD-L1 with a HR of 1.94 and that most of the responders to this treatment had TMB-H [6]. The AtezoTRIBE study found that TMB-H patients would benefit from a greater response rate to a combination of FOLFOXIRI and bevacizumab with or without atezolizumab when compared to low TMB (TMB-L) tumors [7]. In NSCLC and when treated with durvalumab and chemoradiation, patients with TMB-H showed an improved 24-months PFS when compared to TMB-L patients (66% vs. 27%, p = 0.003) [14]. Concordant results have been found in urothelial carcinoma [15], biliary tract cancer [16] and gastric cancer [17]. Keynote 158 is a phase 2 multi-cohort clinical trial that explored TMB expression and its effect on treatment efficacy and response in 10 different cancer types. It concluded that TMB-H was associated with a better outcome when treated with pembrolizumab monotherapy compared to TMB-L. In fact, the objective response rate in the TMB-H group was 29% compared to 6% in the TMB-L group. However, this response rate was assessed by RECIST, which is a standardized method that heavily relies on imaging to assess response to treatment but lacks the general clinical aspect as well as survival rate. In fact, no improvement in OS was reported in this study, underlying the discordance between imaging response and clinical improvement [8]. Keynote 158 is heavily criticized for these reasons, especially since it was the turning point that led to the FDA approval of pembrolizumab usage in TMB-H solid tumors.

Table 1.

Studies in favor of using TMB as an agnostic biomarker for predicting response to immunotherapy.

Clinical Trial/Study	Cancer Type and Studied Therapy	Outcome Measure	In favor of TMB-H	References
Frameshift events predict anti-PD-1/L1 response in head and neck cancer (Hanna et al.)	– Head and neck cancer – Anti – PD-1/L1	Median OS among virus-negative patients	TMB-H (>10) > TMB-L (<5) OS = 20 months > OS = 6 months, respectively	[6]
AtezoTRIBE	– Metastatic colorectal cancer – FOLFOXIRI + Bevacizumab ± Atezolizumab	Median PFS	TMB-H (13.1 months) > TMB-L (11.5 months)	[7]
Keynote 158	– Advanced (unresectable and/or metastatic) solid tumors – Pembrolizumab monotherapy	ORR	TMB-H (ORR = 29%) > TMB-L (ORR = 6%)	[8]
Assessment of Tumor Mutational Burden and Outcomes in Patients With Diverse Advanced Cancers Treated With Immunotherapy (Aggarwal et al.)	– Solid tumors (NSCLC, bladder, head and neck squamous cell cancer …) – Anti-PD-1/PD-L1 ± anti-CTLA-4	OS	TMB-H > TMB-L [HR = 0.72] Prospective subgroup (TMB status evaluated pre ICI administration) TMB-H > TMB-L [HR, 0.61]	[9]
Checkmate 227	– Stage IV NSCLC – Nivolumab + Ipilimumab vs Nivolumab monotherapy or chemotherapy alone	PFS	Nivo + Ipi: TMB-H (7.2 months) > TMB-L (3.2 months)	[10]

Table 2.

Studies against using TMB as an agnostic biomarker for predicting response to immunotherapy.

Clinical Trial/Study	Cancer Type and Studied Therapy	Outcome Measure	Against TMB-H	References
NADIM phase II	– Resectable Stage IIIA NSCLC – Neoadjuvant Nivolumab Plus Chemotherapy	PFS	TMB-H vs TMB-L, [HR = 1.67]	[11]
		OS	TMB-H vs TMB-L, [HR = 2.13]
Tumor mutational load predicts survival after immunotherapy across multiple cancer types (Samstein et al.)	– Glioma – ICI (mostly anti PD-1/PD-L1)	OS	TMB-H vs TMB-L, [HR = 1.3]	[12]
Agenus C-800	– Advanced ovarian cancer – FC-engineered anti-CTLA-4 (Botensilimab) in combination with anti-PDL-1 (Balstilimab)	ORR	Despite over 90% of patients with low TMB, ORR = 33%	[13]
Keynote 158	– Advanced (unresectable and/or metastatic) solid tumors – Pembrolizumab monotherapy	Median OS	TMB-L (12.8 months) > TMB-H (11.7 months)	[8]
Assessment of Tumor Mutational Burden and Outcomes in Patients With Diverse Advanced Cancers Treated With Immunotherapy (Aggarwal et al.)	– Head and neck squamous cell cancer – Anti-PD-1/PD-L1 ± anti-CTLA-4	1 year Survival Probability	TMB-L (60%) > TMB-H (52%)	[9]
Checkmate 227	• Stage IV NSCLC Nivolumab + Ipilimumab vs Nivolumab monotherapy or chemotherapy alone	OS	TMB-H ≡ TMB-L (Nivo+ Ipi VS Chemo; TMB-H HR = 0.68, TMB-L HR = 0.75)	[10]

Another JAMA study found that patients with TMB-H solid tumors, mainly NSCLC, bladder cancer, and head and neck squamous cell carcinoma (HNSCC), had an improved overall survival rate when treated with immune checkpoint inhibitors compared to TMB-L patients (HR 0.74 with upper confidence bound of 0.91). The treatment consisted of a combination of anti-PD-1/PD-L1 and/or anti-CTLA-4 (Ipilimumab, approved in 2020 and tremelimumab, approved in 2022). However, pembrolizumab was by far the most used drug (339 out of 674 patients). In the prospective subgroup, which received the immune checkpoint inhibitor after TMB was evaluated, the OS rate and PFS and TTP were higher in TMB-H patients compared to TMB-L patients in all cancer types studied but not in HNSCC. When focusing on each type of cancer alone, results were also comparable to those in the overall cohort, meaning TMB-H was associated with better OS than TMB-L, except for HNSCC in which the TMB-L population had a better OS than the TMB-H population. Finally, when adjusting to the specific immune checkpoint inhibitor used, the association of TMB-H with a better OS when compared to TMB-L was still valid [9]. This study, on one hand, goes in line with the findings of Keynote 158, as it confirms the positive role of TMB as an agnostic immunotherapy biomarker. However, it contradicts the results of that same study by providing, not only an improvement in radiographic response, but also in clinical aspect and overall survival of patients.

Furthermore, in a letter to NEJM, Rousseau et al. pointed out the limitations of the Keynote 158 study by first stating that, and although it was shown that ICI would have a better response rate based on radiographic response in TMB-H cancers, it did not report an improvement in overall survival rate. Another limitation of this clinical trial that was pointed out was the fact that only 10 rare cancer types were studied while many major tumor types were left out. Thus, in a retrospective cohort studied by Rousseau and his colleagues, it was shown that MMR deficiency, as well as polymerase deficiency (polymerase epsilon and polymerase delta, both referred to as Pol-d) were in fact associated with better response rate to ICI, and not TMB-H in advanced colorectal cancer. Furthermore, and although it has been shown that TMB-H is a good biomarker to predict response to ICI in some MMR proficient cancers (mostly NSCLC, melanoma, and metastatic HNSCC), in many other types, no benefit was seen when MMR and Pol-d were intact. Other than these genetic subgroups (MMR and Pol-d), a subgroup with TMB-H that benefited from a better response to ICI was one that was highly exposed to environmental carcinogens (UV and/or tobacco), which intuitively makes sense since the carcinogens increase the mutagenesis risk. Thus, simply granting FDA approval to use pembrolizumab in TMB-H solid tumors is believed to be a broad criterion and should be reevaluated to be target the most-likely-to-benefit from this therapy patients [18,19].

On the other hand, Provencio reported that the TMB status did not affect the efficiency of nivolumab on resectable stage IIIA NSCLC [11]. It was also reported that, occasionally, low TMB could be associated with a good outcome as well. In fact, when presenting the progress of an ongoing study (Agenus C-800, NCT03860272) about using a new Fc-engineered anti-CTLA4 monoclonal antibody (Botensilimab) in combination with an anti-PD-L1 agent (Balstilimab) in advanced ovarian cancer, it was shown, alongside a favorable outcome in overall response rate, that over 90% of patients had a low TMB, and more than half of them were PD-L1 negative [13]. Surely, we still await the final results of the study to conclude on that, but these preliminary results prove that the controversy over biomarkers; specifically, TMB in this case is very much real and a struggle in the advancement of new treatment options for solid tumors. Another study by Samstein et al. showed that in glioma patients treated by ICI, there was no association between TMB levels and response to treatment. In fact, high levels of TMB were associated with worse response to ICI in this case [12]. Checkmate 227 is a phase 3 clinical trial that studied the efficacy of combining nivolumab with ipilimumab for stage IV NSCLC patients, compared to nivolumab monotherapy or chemotherapy alone. It found that although there was a slightly significant improvement in PFS among TMB-H patients compared to TMB-L patients, there was no significant improvement in OS. It concluded that more data were needed to truly understand the role of TMB as a biomarker [10].

The need to further study the efficacy of TMB as a biomarker is now clear and necessary to improve our understanding of its underlying mechanism and explain the diverging results of different studies.

3. Controversial results among different studies

The discrepancies between different studies and clinical trials can be attributed to different possible causes, illustrated in Figure 2.

Figure 2.

Controversy behind TMB as an agnostic immunotherapy biomarker.

First and foremost, there are some tumor types that are at higher risk of being associated with TMB-H compared to others. For example, in a study published in 2018 and using the Cancer Genome Atlas (TCGA), TMB levels among 24 cancer types were evaluated. Eleven cancer types (melanoma, lung squamous cell carcinoma, lung adenocarcinoma, transitional cell carcinoma, uterus adenocarcinoma, stomach adenocarcinoma, head and neck squamous cell carcinoma, colorectal carcinoma, cervix carcinoma, esophagus adenocarcinoma, and sarcoma) presented at least 9% of TMB-H patients, while the 13 others presented less than 5% of TMB-H patients [20].

Additionally, carcinogen exposure appears to also play an important role in TMB quantification, as tumors that are correlated to a certain carcinogen (such as melanoma and lung cancer) tended to have much higher TMB levels [21]. Young Kwang Chae et al. found that smoking history was associated with higher TMB, further supporting the idea that carcinogen exposition would play a role in higher levels of TMB. This is intuitively accurate, as it is widely accepted that exposition to such substances may induce DNA damage, and consequently, more DNA variations. However, they also found that TMB was independent of the disease stage, which would imply that TMB levels are primarily associated with DNA damage and not with the tumor aggressivity [22]. This was further reinforced by Alexandrov et al. in their paper about signatures of mutational processes in cancer. In fact, it was observed that melanoma and NSCLC often have high TMB due to the mutagenic effect of exposition to UV light and smoking, respectively [23]. This shows that different tumor types not only are correlated with different levels of TMB but that the association between TMB expression and response to ICI might also be related to the tumor type. In addition to absolute TMB levels, some tumor types have been found to have higher variability levels of TMB, such as uterine, colon, and bladder cancer, while lung cancer and HNSCC have less variability [24].

Second, the cutoff definition of TMB and the sequencing assays used to define this value are very different from one institution to another. For example, in a study by Young Kwang Chae et al., the cutoff value defined for TMB-H was 20–34 mutations/Mb, with anything over 34 mutations/Mb being defined as TMB very high [22]. Opposite to that, Tang et al. used a cutoff of 6 mutations/Mb for TMB-H [25]. In the clinical trial Checkmate-026 testing nivolumab on stage IV NSCLC, patients having more than 243 mutations were associated with a better outcome [26]. Additionally, for cancers that have been reported to have lower TMB levels, such as pancreatic cancer, it is intuitively reasonable for researchers like Imamura et al. to test the reduction of the TMB-H cutoff to 5 mutations/Mb [27]. Not only that, but a meta-analysis conducted by Wu et al. showed that some patients with TMB-H benefitted from ICI use compared to TMB-L patients in some cancer types, like NSCLC and melanoma, but not in others like urothelial cancer or gastroesophageal cancer [28]. This could be attributed to a wrong definition of TMB cutoff in certain tumor types. In another study published by Wildsmith et al. evaluating the association of TMB with OS when treating HNSCC with a combination of durvalumab and tremelimumab at different cutoffs, TMB-H was associated with a higher OS and was most significant when using a cutoff of 10 or 12 mutations/Mb [29]. Moreover, the FDA had initially proposed to use a cutoff of 10 mutations/Mb to define TMB-H, and other than the actual arbitrary cutoff, many have criticized this bimodal classification, hoping to replace it with a three-way classification: high, intermediate, and low, paving the way for a gray-zone that could help in future TMB panel designs [4,30]. Imamura et al. also introduced such a classification by defining a low TMB between 1 and 5 mutations/Mb and an ultra-low TMB of less than 1 [27]. Chalmers et al. also proved that the median TMB for pediatric malignancies was twice as low as TMB for adult malignancies [21]. In addition, it was demonstrated that TMB tends to increase with age in most cancer types [31]. All these data further support that we still need a better understanding of the cutoff used to define TMB-H both in general and specifically for each cancer type and age range.

Furthermore, the use of different sequencing panel assays with different genes included does not help unify opinions around the subject. In fact, because whole-exome sequencing is still not a viable option for clinical use, different panels have to be used to minimize cost and time consumption. Thus, many panels have emerged, with most importantly the Memorial Sloan Kettering-Integrated Mutation Profiling of Actionable Cancer Targets (MSK-IMPACT) and Foundation-One CDx; both of which are FDA-approved [32]. This variability makes it harder to obtain reproducible results and, consequently, unified conclusions about the use of TMB as a biomarker [4]. Also, the lack of recommendations on how to sequence a tumor for TMB, differences in the location, size and type of mutations sequenced, and the different panels used all render the cutoff value issue exponentially more complicated [21].

Moreover, the association between TMB and other biomarkers has to be studied. First, many papers have tried to elucidate the association between PD-L1 status and TMB levels. In their study, Aggarwal et al. concluded that it is still unclear if the association between TMB-H and OS is independent of PD-L1 status [9]. Yarchoan et al. found that, apart from a few exceptions, TMB levels and PD-L1 were mostly independent of each other. In fact, in their study, and although there were a few cancer types in which a slight correlation between the two biomarkers was present, no statistically significant association could be concluded upon. When exploring the effect of these results on the response to immune checkpoint inhibitors (anti-PD-1 therapy), patients with TMB-H had a better ORR, regardless of PD-L1 status [33]. This has also been seen in the clinical trial Checkmate-568 which evaluates the efficacy of nivolumab in combination with low-dose ipilimumab when treating advanced NSCLC, as all patients with TMB-H, regardless of PD-L1 status, had a better outcome and improved PFS. However, in patients with TMB-L, the ORR was correlated with the level of PD-L1 expression: the higher the expression of PD-L1 was, the better was the response rate [34]. While this finding further reinforces the proof that TMB-H is associated with better outcomes from ICI therapy, it also proves that there are many tumor types that might not benefit from this increased response if we only test for PD-L1 status. It also shows that reversal of PD-1 immune suppression is a less important mechanism in TMB-H tumors [33]. Furthermore, response to ICI therapy has been found to be affected by other confounding factors like obesity, as well as lifestyle factors like smoking and exercising and sociological factors like sex, race, and age. All these factors further hamper the advancements needed to fully understand the role of TMB as an agnostic immunotherapy biomarker [35]. All this data shows how little true understanding of the underlying mechanisms and of the predictions of response to treatment we have, causing TMB to be such a controversial topic. Still, as long as we do not have a clear definite conclusion on this issue, it is still suggested that both PD-L1 status and TMB levels be used to better predict responsiveness to immune checkpoint inhibitors [36].

Finally, the use of specific composite biomarkers, such as mTMB (modified tumor mutational burden), has also been discussed recently. In a retrospective study published in The Lancet Oncology, the genomic expression of over 48,000 TMB-H MSS gastrointestinal tumors was studied. It was demonstrated that, despite being TMB-H, patients presenting a certain panel of gene mutations in the tumor (including SMAD2, mTOR, and RB1) would present a worse response to immune checkpoint inhibitors compared to patients presenting another panel of gene mutations (including JAK2, MAP2K1, MAP2K4) which would be correlated to a better response to ICIs. This leads us to believe that there might be a genetic signature to tumors that would be more responsive to treatment, and qualifying the mutations inside a tumor might be more useful than quantifying TMB [37].

4. Conclusion and future perspective

Tumor mutation burden has been a topic rising in interest in recent years, being a very promising biomarker for cancer treatment efficacy prediction. However, it is yet to make its breakthrough inside the clinic of oncologists, as it is still a widely controversial biomarker. In fact, studies like AtezoTRIBE and Keynote 158 underline the positive role of TMB as an agnostic ICI predictive biomarker, while other studies like Checkmate 227 and Agenus C-800 show no correlation, or even negative correlation between TMB levels and ICI response rate. Not only that, but some studies only rely on radiographic results, like Keynote 158, which is arguably of lesser interest if no OS or PFS improvement is reported. All these discrepancies are partly attributed to the complexity of TMB as a biomarker, variations in different patient populations, and treatments used in studies, but also due to bad cutoff definition and limited knowledge about its relationship with other biomarkers such as the PD-L1 status, as well as other confounding factors like diabetes and metabolic genes, sex, and race. Furthermore, the different immune checkpoint inhibitors, as well as other forms of immunotherapy that have been emerging recently, such as vaccines, complicate even more the study of TMB as an agnostic predictive biomarker. Consequently, waiting for more comprehensive data on the TMB levels among different patient populations, ages, and cancer types is crucial to set a more accurate cutoff definition, which could be then used to study the relationship of TMB with other biomarkers and its efficacy in predicting response to treatment. Another aspect that should be explored is the genetic signature of tumors in order to truly understand the value of TMB, in its genetic mutations qualitatively and not quantitatively, which is looking to be the most promising use of TMB. In fact, tumor heterogeneity, alongside all the many factors already mentioned, from patient characteristics to different tumor types and sequencing assays, makes it unlikely that a single quantitative TMB cutoff could be used as a predictor of immunotherapy response poor. However, finding genetic signatures that could predict immunotherapy response looks more promising and makes a good argument for a modified qualitative TMB. To achieve all this and finally use, or not, TMB in the clinical setting as an agnostic immunotherapy predictive biomarker, different studies should focus on TMB as a primary endpoint, which has scarcely been done to date. In fact, in most studies and papers published in the scientific community, the role of TMB as a biomarker is either a secondary endpoint or a post-hoc analysis that was obtained coincidentally. However, to advance, clinical oncologists and researchers must put TMB front and center to unify results about the discrepancies and underlying mechanisms of this promising biomarker.

Funding Statement

This paper was not funded.

Article highlights

• TMB is a promising agnostic immunotherapy biomarker but diverging results in theliterature hamper its use in our everyday practice.

• Different studies underline the positive role of TMB as an agnostic biomarker, whileothers don’t find a role, or even a negative role of TMB.

• The use of different cutoff levels for TMB definitions, different sequencing assays,different tumor types and other confounding factors can be attributed to thedivergence of results.

• The use of qualitative TMB might be more interesting than quantitative TMB, in thescope of understanding the genomic signature of a tumor.

• Exploring the role of TMB as a primary endpoint in trials and studies is essential inunifying our findings, regardless of the expected results.

Authors contributions

Antoine Mouawad: Writing – First Draft, Writing – Editing and Review, Project Conceptualization, Data Collection, and Data Analysis.

Marc Boutros: Writing – Editing and Review, Project Conceptualization, and Data Analysis.

Antoine Chartouni: Writing – Editing and Review, Project Conceptualization, Data Analysis, and Figures and Tables Drafting.

Fouad Attieh: Writing – Editing and Review, Project Conceptualization, and Data Analysis.

Hampig Raphaël Kourie: Project Administration, Supervision, Writing – Editing and Review, and Project Conceptualization.

Disclosure statement

The authors have no relevant affiliations or financial involvement with any organization or entity with a financial interest in or financial conflict with the subject matter or materials discussed in the manuscript. This includes employment, consultancies, honoraria, stock ownership or options, expert testimony, grants or patents received or pending, or royalties.

No writing assistance was utilized in the production of this manuscript.